{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Actions as vector, and RL agent training"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It is recommended to have a look at the [0_basic_functionalities](0_basic_functionalities.ipynb), [1_Observation_Agents](1_Observation_Agents.ipynb) and [2_Action_GridManipulation](2_Action_GridManipulation.ipynb) notebooks before getting into this one."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Objectives**\n",
    "\n",
    "In this notebook we will expose :\n",
    "* how to use the \"converters\": some specific action_space that allows to manipulate a specific action representation\n",
    "* how to train a (stupid) Agent using reinforcement learning.\n",
    "* how to inspect (rapidly) the action taken by the Agent\n",
    "\n",
    "**NB** for this tutorial we train an Agent inspired from this blog post: [deep-reinforcement-learning-tutorial-with-open-ai-gym](https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368). Many other different reinforcement learning tutorial exist. The code showed in this notebook has no pretention except to demonstrate how to use Grid2Op functionality to train a Deep Reinforcement learning Agent and inspect its behaviour. There are absolutely nothing implied about the performance, training strategy, type of Agent, meta parameters etc. All of them are purely \"random\".\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import sys\n",
    "import grid2op"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Impossible to automatically add a menu / table of content to this notebook.\n",
      "You can download \"jyquickhelper\" package with: \n",
      "\"pip install jyquickhelper\"\n"
     ]
    }
   ],
   "source": [
    "res = None\n",
    "try:\n",
    "    from jyquickhelper import add_notebook_menu\n",
    "    res = add_notebook_menu()\n",
    "except ModuleNotFoundError:\n",
    "    print(\"Impossible to automatically add a menu / table of content to this notebook.\\nYou can download \\\"jyquickhelper\\\" package with: \\n\\\"pip install jyquickhelper\\\"\")\n",
    "res"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## I) Manipulating action representation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Grid2op package has been built with an \"object oriented\" perspective: almost everything is encapsulated in a dedicated `class`. This allows for more customization of the plateform.\n",
    "\n",
    "The downside of this approach is that machine learning method, and especially deep learning, often prefers to deal with vectors rather than with `complex` objects. Indeed, as we covered in the previous tutorials on the platform, building our own actions can be tedious and can sometime require knowledge of the powergrid.\n",
    "\n",
    "On the contrary, in most of standard Reinforcement Learning environment, actions have an higher representation. For example in pacman, there are 4 different types of actions: turn left, turn right, go up or do down. This allows for easy sampling (you need to achieve a uniform sampling you simply need to sample a number between 0 and 3 included) and an easy representation: each action is a different component of a vector of dimension 4 [because there are 4 actions]. \n",
    "\n",
    "On the other hand this representation is not \"human friendly\". It is quite convenient in the case of pacman because the action space is rather small making it possible to remember which action corresponds to which component, but in the case of the grid2op package, there are hundreds, sometimes thousands of actions, making it impossible to remember which component corresponds to which actions. We suppose we don't really care about this fact here, as tutorials on Reinforcement Learning with discrete action space often assume that actions are labelled with integer (such as in pacman for example).\n",
    "\n",
    "Howerever, to allow the training of RL agent more easily, we allows to make some \"[Converters](https://grid2op.readthedocs.io/en/latest/converters.html)\" whose roles are to allow an agent to deal with a custom representation of the action space. The class [AgentWithConverter](https://grid2op.readthedocs.io/en/latest/agent.html#grid2op.Agent.AgentWithConverter) is perfect for such usage."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/benjamin/Documents/grid2op_test/getting_started/grid2op/MakeEnv.py:668: UserWarning:\n",
      "\n",
      "Your are using only 2 chronics for this environment. More can be download by running, from a command line:\n",
      "python -m grid2op.download --name \"case14_redisp\" --path_save PATH\\WHERE\\YOU\\WANT\\TO\\DOWNLOAD\\DATA\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# import the usefull class\n",
    "import numpy as np\n",
    "\n",
    "from grid2op import make\n",
    "from grid2op.Agent import RandomAgent \n",
    "max_iter = 100 # to make computation much faster we will only consider 50 time steps instead of 287\n",
    "\n",
    "env = make(name_env=\"case14_redisp\")\n",
    "env.seed(0)  # this is to ensure the same action are taken by the \"RandomAgent\".\n",
    "my_agent =  RandomAgent(env.action_space)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And that's it. This agent will be able to perform any action, but instead of going through the description of the actions from a powersystem point of view (ie setting what is connected to what, what is disconnected etc.) it will simply choose an integer with the method `my_act` this integer will then be converter back to a proper valid action.\n",
    "\n",
    "Here we have an example on the action representation as seen by the Agent:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "172\n",
      "47\n",
      "117\n"
     ]
    }
   ],
   "source": [
    "for el in range(3):\n",
    "    print(my_agent.my_act(None, None))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And below you can see the \"`act`\" functions behaves as expected:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "This action will:\n",
      "\t - NOT change anything to the injections\n",
      "\t - NOT perform any redispatching action\n",
      "\t - NOT force any line status\n",
      "\t - NOT switch any line status\n",
      "\t - NOT switch anything in the topology\n",
      "\t - Set the bus of the following element:\n",
      "\t \t - assign bus 2 to line (origin) 12 [on substation 5]\n",
      "\t \t - assign bus 2 to line (origin) 13 [on substation 5]\n",
      "\t \t - assign bus 1 to line (origin) 14 [on substation 5]\n",
      "\t \t - assign bus 2 to line (extremity) 17 [on substation 5]\n",
      "\t \t - assign bus 1 to generator 2 [on substation 5]\n",
      "\t \t - assign bus 1 to load 5 [on substation 5]\n",
      "This action will:\n",
      "\t - NOT change anything to the injections\n",
      "\t - NOT perform any redispatching action\n",
      "\t - NOT force any line status\n",
      "\t - NOT switch any line status\n",
      "\t - Change the bus of the following element:\n",
      "\t \t - switch bus of line (origin) 3 [on substation 8]\n",
      "\t \t - switch bus of line (extremity) 16 [on substation 8]\n",
      "\t \t - switch bus of line (extremity) 19 [on substation 8]\n",
      "\t - NOT force any particular bus configuration\n",
      "This action will:\n",
      "\t - NOT change anything to the injections\n",
      "\t - NOT perform any redispatching action\n",
      "\t - NOT force any line status\n",
      "\t - NOT switch any line status\n",
      "\t - Change the bus of the following element:\n",
      "\t \t - switch bus of generator 1 [on substation 2]\n",
      "\t \t - switch bus of load 1 [on substation 2]\n",
      "\t - NOT force any particular bus configuration\n"
     ]
    }
   ],
   "source": [
    "for el in range(3):\n",
    "    print(my_agent.act(None, None))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**NB** lots of these actions are equivalent to the \"do nothing\" action at some point. For example, when trying to reconnect a powerline that is already connected will do nothing. Same for the topology. If everything is already connected to bus 1, then the action to connect things to bus 1 on the same substation will not affect the powergrid."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## II) Training an Agent"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For this tutorial, we will expose to built a Q-learning Agent. Most of the code (except the neural network architecture) are inspired from this blog post: [https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368](https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368).\n",
    "\n",
    "**Requirements** This notebook require to have `keras` installed on your machine.\n",
    "\n",
    "As always in these notebook, we will use the `case14_fromfile` Environment. No proper care has been taken to set the thermal limits on this grid. It's unlikely that the agent can learn anything in this context."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### II.A) Defining some \"helpers\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The type of Agent were are using require a bit of set up, independantly of Grid2Op. We will reuse the code showed in \n",
    "[https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368](https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368) and in [Reinforcement-Learning-Tutorial](https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial) from Abhinav Sagar code under a *MIT license* found here: [MIT License](https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial/blob/master/LICENSE).\n",
    "\n",
    "This first section is here to define these classes."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "But first let's import the necessary dependencies"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "#tf2.0 friendly\n",
    "import numpy as np\n",
    "import random\n",
    "import warnings\n",
    "with warnings.catch_warnings():\n",
    "    warnings.filterwarnings(\"ignore\", category=FutureWarning)\n",
    "    import tensorflow.keras\n",
    "    import tensorflow.keras.backend as K\n",
    "    from tensorflow.keras.models import load_model, Sequential, Model\n",
    "    from tensorflow.keras.optimizers import Adam\n",
    "    from tensorflow.keras.layers import Activation, Dropout, Flatten, Dense, subtract, add\n",
    "    from tensorflow.keras.layers import Input, Lambda, Concatenate"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### a) Replay buffer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    " First we define a \"replay buffer\" necessary to train the Agent."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Credit Abhinav Sagar: \n",
    "# https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial\n",
    "# Code under MIT license, available at:\n",
    "# https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial/blob/master/LICENSE\n",
    "from collections import deque\n",
    "\n",
    "class ReplayBuffer:\n",
    "    \"\"\"Constructs a buffer object that stores the past moves\n",
    "    and samples a set of subsamples\"\"\"\n",
    "\n",
    "    def __init__(self, buffer_size):\n",
    "        self.buffer_size = buffer_size\n",
    "        self.count = 0\n",
    "        self.buffer = deque()\n",
    "\n",
    "    def add(self, s, a, r, d, s2):\n",
    "        \"\"\"Add an experience to the buffer\"\"\"\n",
    "        # S represents current state, a is action,\n",
    "        # r is reward, d is whether it is the end, \n",
    "        # and s2 is next state\n",
    "        experience = (s, a, r, d, s2)\n",
    "        if self.count < self.buffer_size:\n",
    "            self.buffer.append(experience)\n",
    "            self.count += 1\n",
    "        else:    \n",
    "            self.buffer.popleft()\n",
    "            self.buffer.append(experience)\n",
    "\n",
    "    def size(self):\n",
    "        return self.count\n",
    "\n",
    "    def sample(self, batch_size):\n",
    "        \"\"\"Samples a total of elements equal to batch_size from buffer\n",
    "        if buffer contains enough elements. Otherwise return all elements\"\"\"\n",
    "\n",
    "        batch = []\n",
    "\n",
    "        if self.count < batch_size:\n",
    "            batch = random.sample(self.buffer, self.count)\n",
    "        else:\n",
    "            batch = random.sample(self.buffer, batch_size)\n",
    "\n",
    "        # Maps each experience in batch in batches of states, actions, rewards\n",
    "        # and new states\n",
    "        s_batch, a_batch, r_batch, d_batch, s2_batch = list(map(np.array, list(zip(*batch))))\n",
    "\n",
    "        return s_batch, a_batch, r_batch, d_batch, s2_batch\n",
    "\n",
    "    def clear(self):\n",
    "        self.buffer.clear()\n",
    "        self.count = 0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### b) Meta parameters of the methods"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Then we re-use the default parameters, note that these can be optimized. Nothing has been changed for this example.\n",
    "\n",
    "For more information about them, please refer to the blog post of Abhinav Sagar [available here](https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "DECAY_RATE = 0.9\n",
    "BUFFER_SIZE = 40000\n",
    "MINIBATCH_SIZE = 64\n",
    "TOT_FRAME = 3000000\n",
    "EPSILON_DECAY = 10000\n",
    "MIN_OBSERVATION = 42 #5000\n",
    "FINAL_EPSILON = 1/300  # have on average 1 random action per scenario of approx 287 time steps\n",
    "INITIAL_EPSILON = 0.1\n",
    "TAU = 0.01\n",
    "ALPHA = 1\n",
    "# Number of frames to \"throw\" into network\n",
    "NUM_FRAMES = 1 ## this has been changed compared to the original implementation."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### II.B) Adapatation of the inputs"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the original code, the models were used to play an Atari game and the inputs were images. For our system, the inputs are \"Observation\" converted as vector.\n",
    "\n",
    "For a more detailed description of the code used, please check:\n",
    "* [https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368](https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368)\n",
    "* and [Reinforcement-Learning-Tutorial](https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial)\n",
    "\n",
    "\n",
    "This is why we adapted the original code from Abhinav Sagar:\n",
    "* We replaced convolutional layers with fully connected (dense) layers\n",
    "* We made sure not to look at all the observations, but rather at only some part of it."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### a) extracting relevant information of observation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First we extract relevant information about the dimension of the observation space, and the action space."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "OBSERVATION_SIZE = env.observation_space.size()\n",
    "NUM_ACTIONS = my_agent.action_space.n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A few comments here.\n",
    "\n",
    "First, we don't change anything to the observation space. This means that the vector the agent will receive is really big, not scaled and with lots of informations that are not really usefull."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### b) Code the neural networks"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The code of the neural networks, used to solve this problem is modified very slightly to adapt. The biggest changes come from removing the convolutional layers, as well as adapting the input and output size.\n",
    "\n",
    "For each of the method below, we specify what have been adapted/modified in comparison with the original blog post by Abhinav."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We want to emphasize here that these models are used through the \"`predict_movement`\" method. This method outputs an integer: the action id that maximizes the expected reward at the current tim step in an episode of the game during the training process. It's perfectly suited to use a representation of actions with integer rather than with complete descriptions of what the agent is doing."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### II.C) Making the code of the Agent and train it"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the \"reference\" article [https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368](https://towardsdatascience.com/deep-reinforcement-learning-tutorial-with-open-ai-gym-c0de4471f368), the author Abhinav Sagar made a dedicated environment based on SpaceInvader in the gym repository. We proceed here on a similar way, but with a the grid2op environment."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### a) Adapated code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Credit Abhinav Sagar: \n",
    "# https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial\n",
    "# Code under MIT license, available at:\n",
    "# https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial/blob/master/LICENSE\n",
    "\n",
    "class DeepQ(object):\n",
    "    \"\"\"Constructs the desired deep q learning network\"\"\"\n",
    "    def __init__(self, action_size, lr=1e-5, observation_size=OBSERVATION_SIZE):\n",
    "        # It is not modified from  Abhinav Sagar's code, except for adding the possibility to change the learning rate\n",
    "        # in parameter is also present the size of the action space\n",
    "        # (it used to be a global variable in the original code)\n",
    "        self.action_size = action_size\n",
    "        self.observation_size = observation_size\n",
    "        self.model = None\n",
    "        self.target_model = None\n",
    "        self.lr_ = lr\n",
    "        self.qvalue_evolution = []\n",
    "        self.construct_q_network()\n",
    "    \n",
    "    def construct_q_network(self):\n",
    "        # replacement of the Convolution layers by Dense layers, and change the size of the input space and output space\n",
    "        \n",
    "        # Uses the network architecture found in DeepMind paper\n",
    "        self.model = Sequential()\n",
    "        input_layer = Input(shape = (self.observation_size*NUM_FRAMES,))\n",
    "        layer1 = Dense(self.observation_size*NUM_FRAMES)(input_layer)\n",
    "        layer1 = Activation('relu')(layer1)\n",
    "        layer2 = Dense(self.observation_size)(layer1)\n",
    "        layer2 = Activation('relu')(layer2)\n",
    "        layer3 = Dense(self.observation_size)(layer2)\n",
    "        layer3 = Activation('relu')(layer3)\n",
    "        layer4 = Dense(2*NUM_ACTIONS)(layer3)\n",
    "        layer4 = Activation('relu')(layer4)\n",
    "        output = Dense(NUM_ACTIONS)(layer4)\n",
    "        \n",
    "        self.model = Model(inputs=[input_layer], outputs=[output])\n",
    "        self.model.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "        self.target_model = Model(inputs=[input_layer], outputs=[output])\n",
    "        self.target_model.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "        self.target_model.set_weights(self.model.get_weights())\n",
    "    \n",
    "    def predict_movement(self, data, epsilon):\n",
    "        \"\"\"Predict movement of game controler where is epsilon\n",
    "        probability randomly move.\"\"\"\n",
    "        # nothing has changed from the original implementation\n",
    "        rand_val = np.random.random()\n",
    "        q_actions = self.model.predict(data.reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "        \n",
    "        if rand_val < epsilon:\n",
    "            opt_policy = np.random.randint(0, NUM_ACTIONS)\n",
    "        else:\n",
    "            opt_policy = np.argmax(np.abs(q_actions))\n",
    "            \n",
    "        self.qvalue_evolution.append(q_actions[0,opt_policy])\n",
    "\n",
    "        return opt_policy, q_actions[0, opt_policy]\n",
    "\n",
    "    def train(self, s_batch, a_batch, r_batch, d_batch, s2_batch, observation_num):\n",
    "        \"\"\"Trains network to fit given parameters\"\"\"\n",
    "        # nothing has changed from the original implementation, except for changing the input dimension 'reshape'\n",
    "        batch_size = s_batch.shape[0]\n",
    "        targets = np.zeros((batch_size, NUM_ACTIONS))\n",
    "\n",
    "        for i in range(batch_size):\n",
    "            targets[i] = self.model.predict(s_batch[i].reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "            fut_action = self.target_model.predict(s2_batch[i].reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "            targets[i, a_batch[i]] = r_batch[i]\n",
    "            if d_batch[i] == False:\n",
    "                targets[i, a_batch[i]] += DECAY_RATE * np.max(fut_action)\n",
    "        loss = self.model.train_on_batch(s_batch, targets)\n",
    "        # Print the loss every 100 iterations.\n",
    "        if observation_num % 100 == 0:\n",
    "            print(\"We had a loss equal to \", loss)\n",
    "\n",
    "    def save_network(self, path):\n",
    "        # Saves model at specified path as h5 file\n",
    "        # nothing has changed\n",
    "        self.model.save(path)\n",
    "        print(\"Successfully saved network.\")\n",
    "\n",
    "    def load_network(self, path):\n",
    "        # nothing has changed\n",
    "        self.model = load_model(path)\n",
    "        print(\"Succesfully loaded network.\")\n",
    "\n",
    "    def target_train(self):\n",
    "        # nothing has changed from the original implementation\n",
    "        model_weights = self.model.get_weights()\n",
    "        target_model_weights = self.target_model.get_weights()\n",
    "        for i in range(len(model_weights)):\n",
    "            target_model_weights[i] = TAU * model_weights[i] + (1 - TAU) * target_model_weights[i]\n",
    "        self.target_model.set_weights(target_model_weights)\n",
    "        \n",
    "class DuelQ(object):\n",
    "    \"\"\"Constructs the desired deep q learning network\"\"\"\n",
    "    def __init__(self, action_size, lr=0.00001, observation_size=OBSERVATION_SIZE):\n",
    "        # It is not modified from  Abhinav Sagar's code, except for adding the possibility to change the learning rate\n",
    "        # in parameter is also present the size of the action space\n",
    "        # (it used to be a global variable in the original code)\n",
    "        self.action_size = action_size\n",
    "        self.observation_size = observation_size\n",
    "        self.lr_ = lr\n",
    "        self.model = None\n",
    "        self.qvalue_evolution = []\n",
    "        self.construct_q_network()\n",
    "\n",
    "    def construct_q_network(self):\n",
    "        # Uses the network architecture found in DeepMind paper\n",
    "        # The inputs and outputs size have changed, as well as replacing the convolution by dense layers.\n",
    "        self.model = Sequential()\n",
    "        \n",
    "        input_layer = Input(shape = (self.observation_size*NUM_FRAMES,))\n",
    "        lay1 = Dense(self.observation_size*NUM_FRAMES)(input_layer)\n",
    "        lay1 = Activation('relu')(lay1)\n",
    "        \n",
    "        lay2 = Dense(self.observation_size)(lay1)\n",
    "        lay2 = Activation('relu')(lay2)\n",
    "        \n",
    "        lay3 = Dense(2*NUM_ACTIONS)(lay2)\n",
    "        lay3 = Activation('relu')(lay3)\n",
    "        \n",
    "        fc1 = Dense(NUM_ACTIONS)(lay3)\n",
    "        advantage = Dense(NUM_ACTIONS)(fc1)\n",
    "        fc2 = Dense(NUM_ACTIONS)(lay3)\n",
    "        value = Dense(1)(fc2)\n",
    "        \n",
    "        meaner = Lambda(lambda x: K.mean(x, axis=1) )\n",
    "        mn_ = meaner(advantage)  \n",
    "        tmp = subtract([advantage, mn_])  # keras doesn't like this part...\n",
    "        policy = add([tmp, value])\n",
    "\n",
    "        self.model = Model(inputs=[input_layer], outputs=[policy])\n",
    "        self.model.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "\n",
    "        self.target_model = Model(inputs=[input_layer], outputs=[policy])\n",
    "        self.target_model.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "        print(\"Successfully constructed networks.\")\n",
    "    \n",
    "    def predict_movement(self, data, epsilon):\n",
    "        \"\"\"Predict movement of game controler where is epsilon\n",
    "        probability randomly move.\"\"\"\n",
    "        # only changes lie in adapting the input shape\n",
    "        q_actions = self.model.predict(data.reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "        opt_policy = np.argmax(q_actions)\n",
    "        rand_val = np.random.random()\n",
    "        if rand_val < epsilon:\n",
    "            opt_policy = np.random.randint(0, NUM_ACTIONS)\n",
    "            \n",
    "        self.qvalue_evolution.append(q_actions[0,opt_policy])\n",
    "\n",
    "        return opt_policy, q_actions[0, opt_policy]\n",
    "\n",
    "    def train(self, s_batch, a_batch, r_batch, d_batch, s2_batch, observation_num):\n",
    "        \"\"\"Trains network to fit given parameters\"\"\"\n",
    "        # nothing has changed except adapting the input shapes\n",
    "        batch_size = s_batch.shape[0]\n",
    "        targets = np.zeros((batch_size, NUM_ACTIONS))\n",
    "\n",
    "        for i in range(batch_size):\n",
    "            targets[i] = self.model.predict(s_batch[i].reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "            fut_action = self.target_model.predict(s2_batch[i].reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "            targets[i, a_batch[i]] = r_batch[i]\n",
    "            if d_batch[i] == False:\n",
    "                targets[i, a_batch[i]] += DECAY_RATE * np.max(fut_action)\n",
    "\n",
    "        loss = self.model.train_on_batch(s_batch, targets)\n",
    "\n",
    "        # Print the loss every 100 iterations.\n",
    "        if observation_num % 100 == 0:\n",
    "            print(\"We had a loss equal to \", loss)\n",
    "\n",
    "    def save_network(self, path):\n",
    "        # Saves model at specified path as h5 file\n",
    "        # nothing has changed\n",
    "        self.model.save(path)\n",
    "        print(\"Successfully saved network.\")\n",
    "\n",
    "    def load_network(self, path):\n",
    "        # nothing has changed\n",
    "        self.model.load_weights(path)\n",
    "        self.target_model.load_weights(path)\n",
    "        print(\"Succesfully loaded network.\")\n",
    "\n",
    "    def target_train(self):\n",
    "        # nothing has changed\n",
    "        model_weights = self.model.get_weights()\n",
    "        self.target_model.set_weights(model_weights)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Another custom made version of the Q-Learning algorithm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "class RealQ(object):\n",
    "    \"\"\"Constructs the desired deep q learning network\"\"\"\n",
    "    def __init__(self, action_size, lr=1e-5, observation_size=OBSERVATION_SIZE, mean_reg=False):\n",
    "\n",
    "        self.action_size = action_size\n",
    "        self.observation_size = observation_size\n",
    "        self.model = None\n",
    "        self.target_model = None\n",
    "        self.lr_ = lr\n",
    "        self.mean_reg = mean_reg\n",
    "        \n",
    "        self.qvalue_evolution=[]\n",
    "\n",
    "        self.construct_q_network()\n",
    "        \n",
    "    def construct_q_network(self):\n",
    "\n",
    "        self.model = Sequential()\n",
    "        \n",
    "        input_states = Input(shape = (self.observation_size,))\n",
    "        input_action = Input(shape = (NUM_ACTIONS,))\n",
    "        input_layer = Concatenate()([input_states, input_action])\n",
    "        \n",
    "        lay1 = Dense(self.observation_size)(input_layer)\n",
    "        lay1 = Activation('relu')(lay1)\n",
    "        \n",
    "        lay2 = Dense(self.observation_size)(lay1)\n",
    "        lay2 = Activation('relu')(lay2)\n",
    "        \n",
    "        lay3 = Dense(2*NUM_ACTIONS)(lay2)\n",
    "        lay3 = Activation('relu')(lay3)\n",
    "        \n",
    "        fc1 = Dense(NUM_ACTIONS)(lay3)\n",
    "        advantage = Dense(1, activation = 'linear')(fc1)\n",
    "        \n",
    "        if self.mean_reg==True:\n",
    "            advantage = Lambda(lambda x : x - K.mean(x))(advantage)\n",
    "        \n",
    "        self.model = Model(inputs=[input_states, input_action], outputs=[advantage])\n",
    "        self.model.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "        \n",
    "        self.model_copy = Model(inputs=[input_states, input_action], outputs=[advantage])\n",
    "        self.model_copy.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "        self.model_copy.set_weights(self.model.get_weights())\n",
    "        \n",
    "    def predict_movement(self, states, epsilon):\n",
    "        \"\"\"Predict movement of game controler where is epsilon\n",
    "        probability randomly move.\"\"\"\n",
    "        # nothing has changed from the original implementation\n",
    "        rand_val = np.random.random()\n",
    "        q_actions = self.model.predict([np.tile(states.reshape(1, self.observation_size),(NUM_ACTIONS,1)), np.eye(NUM_ACTIONS)]).reshape(1,-1)\n",
    "        if rand_val < epsilon:\n",
    "            opt_policy = np.random.randint(0, NUM_ACTIONS)\n",
    "        else:\n",
    "            opt_policy = np.argmax(np.abs(q_actions))\n",
    "            \n",
    "        self.qvalue_evolution.append(q_actions[0,opt_policy])\n",
    "        \n",
    "        return opt_policy, q_actions[0,opt_policy]\n",
    "\n",
    "    def train(self, s_batch, a_batch, r_batch, d_batch, s2_batch, observation_num):\n",
    "        \"\"\"Trains network to fit given parameters\"\"\"\n",
    "        # nothing has changed from the original implementation, except for changing the input dimension 'reshape'\n",
    "        batch_size = s_batch.shape[0]\n",
    "        targets = np.zeros(batch_size)\n",
    "        last_action=np.zeros((batch_size, NUM_ACTIONS))\n",
    "        for i in range(batch_size):\n",
    "            last_action[i,a_batch[i]] = 1\n",
    "            q_pre = self.model.predict([s_batch[i].reshape(1, self.observation_size), last_action[i].reshape(1,-1)], batch_size=1).reshape(1,-1)\n",
    "            q_fut = self.model_copy.predict([np.tile(s2_batch[i].reshape(1, self.observation_size),(NUM_ACTIONS,1)), np.eye(NUM_ACTIONS)]).reshape(1,-1)\n",
    "            fut_action = np.max(q_fut)\n",
    "            if d_batch[i] == False:\n",
    "                targets[i] = ALPHA * (r_batch[i] + DECAY_RATE * fut_action - q_pre)\n",
    "            else:\n",
    "                targets[i] = ALPHA * (r_batch[i] - q_pre)\n",
    "        loss = self.model.train_on_batch([s_batch, last_action], targets)\n",
    "        # Print the loss every 100 iterations.\n",
    "        if observation_num % 100 == 0:\n",
    "            print(\"We had a loss equal to \", loss)\n",
    "\n",
    "    def save_network(self, path):\n",
    "        # Saves model at specified path as h5 file\n",
    "        # nothing has changed\n",
    "        self.model.save(path)\n",
    "        print(\"Successfully saved network.\")\n",
    "\n",
    "    def load_network(self, path):\n",
    "        # nothing has changed\n",
    "        self.model = load_model(path)\n",
    "        print(\"Succesfully loaded network.\")\n",
    "\n",
    "    def target_train(self):\n",
    "        # nothing has changed from the original implementation\n",
    "        model_weights = self.model.get_weights() \n",
    "        target_model_weights = self.model_copy.get_weights()\n",
    "        for i in range(len(model_weights)):\n",
    "            target_model_weights[i] = TAU * model_weights[i] + (1 - TAU) * target_model_weights[i]\n",
    "        self.model_copy.set_weights(model_weights)\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A version of SAC algorithm as described in https://spinningup.openai.com/en/latest/algorithms/sac.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "class SAC(object):\n",
    "    \"\"\"Constructs the desired deep q learning network\"\"\"\n",
    "    def __init__(self, action_size, lr=1e-5, observation_size=OBSERVATION_SIZE):\n",
    "\n",
    "        self.action_size = action_size\n",
    "        self.observation_size = observation_size\n",
    "        self.model = None\n",
    "        self.target_model = None\n",
    "        self.lr_ = lr\n",
    "        self.average_reward = 0\n",
    "        self.life_spent = 1\n",
    "        self.qvalue_evolution = []\n",
    "        self.Is_nan = False\n",
    "\n",
    "        self.construct_q_network()\n",
    "        \n",
    "    def build_q_NN(self):\n",
    "        model = Sequential()\n",
    "        \n",
    "        input_states = Input(shape = (self.observation_size,))\n",
    "        input_action = Input(shape = (NUM_ACTIONS,))\n",
    "        input_layer = Concatenate()([input_states, input_action])\n",
    "        \n",
    "        lay1 = Dense(self.observation_size)(input_layer)\n",
    "        lay1 = Activation('relu')(lay1)\n",
    "        \n",
    "        lay2 = Dense(self.observation_size)(lay1)\n",
    "        lay2 = Activation('relu')(lay2)\n",
    "        \n",
    "        lay3 = Dense(2*NUM_ACTIONS)(lay2)\n",
    "        lay3 = Activation('relu')(lay3)\n",
    "        \n",
    "        advantage = Dense(1, activation = 'linear')(lay3)\n",
    "        \n",
    "        model = Model(inputs=[input_states, input_action], outputs=[advantage])\n",
    "        model.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "        \n",
    "        return model\n",
    "    \n",
    "    def construct_q_network(self):\n",
    "        #construct double Q networks\n",
    "        self.model_Q = self.build_q_NN()\n",
    "        self.model_Q2 = self.build_q_NN()\n",
    "\n",
    "    \n",
    "        #state value function approximation\n",
    "        self.model_value = Sequential()\n",
    "        \n",
    "        input_states = Input(shape = (self.observation_size,))\n",
    "        \n",
    "        lay1 = Dense(self.observation_size)(input_states)\n",
    "        lay1 = Activation('relu')(lay1)\n",
    "        \n",
    "        lay3 = Dense(2*NUM_ACTIONS)(lay1)\n",
    "        lay3 = Activation('relu')(lay3)\n",
    "        \n",
    "        advantage = Dense(NUM_ACTIONS, activation = 'relu')(lay3)\n",
    "        state_value = Dense(1, activation = 'linear')(advantage)\n",
    "        \n",
    "        self.model_value = Model(inputs=[input_states], outputs=[state_value])\n",
    "        self.model_value.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "        \n",
    "        self.model_value_target = Sequential()\n",
    "        \n",
    "        input_states = Input(shape = (self.observation_size,))\n",
    "        \n",
    "        lay1 = Dense(self.observation_size)(input_states)\n",
    "        lay1 = Activation('relu')(lay1)\n",
    "        \n",
    "        lay3 = Dense(2*NUM_ACTIONS)(lay1)\n",
    "        lay3 = Activation('relu')(lay3)\n",
    "        \n",
    "        advantage = Dense(NUM_ACTIONS, activation = 'relu')(lay3)\n",
    "        state_value = Dense(1, activation = 'linear')(advantage)\n",
    "        \n",
    "        self.model_value_target = Model(inputs=[input_states], outputs=[state_value])\n",
    "        self.model_value_target.compile(loss='mse', optimizer=Adam(lr=self.lr_))\n",
    "        self.model_value_target.set_weights(self.model_value.get_weights())\n",
    "        #policy function approximation\n",
    "        \n",
    "        self.model_policy = Sequential()\n",
    "        input_states = Input(shape = (self.observation_size,))\n",
    "        lay1 = Dense(self.observation_size)(input_states)\n",
    "        lay1 = Activation('relu')(lay1)\n",
    "        \n",
    "        lay2 = Dense(self.observation_size)(lay1)\n",
    "        lay2 = Activation('relu')(lay2)\n",
    "        \n",
    "        lay3 = Dense(2*NUM_ACTIONS)(lay2)\n",
    "        lay3 = Activation('relu')(lay3)\n",
    "        \n",
    "        soft_proba = Dense(NUM_ACTIONS, activation=\"softmax\", kernel_initializer='uniform')(lay3)\n",
    "        \n",
    "        self.model_policy = Model(inputs=[input_states], outputs=[soft_proba])\n",
    "        self.model_policy.compile(loss='categorical_crossentropy', optimizer=Adam(lr=self.lr_))\n",
    "        \n",
    "        print(\"Successfully constructed networks.\")\n",
    "        \n",
    "    def predict_movement(self, states, epsilon):\n",
    "        \"\"\"Predict movement of game controler where is epsilon\n",
    "        probability randomly move.\"\"\"\n",
    "        # nothing has changed from the original implementation\n",
    "        p_actions = self.model_policy.predict(states.reshape(1, self.observation_size)).ravel()\n",
    "        rand_val = np.random.random()\n",
    "        \n",
    "        if rand_val < epsilon / 10:\n",
    "            opt_policy = np.random.randint(0, NUM_ACTIONS)\n",
    "        else:\n",
    "            #opt_policy = np.random.choice(np.arange(NUM_ACTIONS, dtype=int), size=1, p = p_actions)\n",
    "            opt_policy = np.argmax(p_actions)\n",
    "                            \n",
    "        return np.int(opt_policy), p_actions[opt_policy]\n",
    "\n",
    "    def train(self, s_batch, a_batch, r_batch, d_batch, s2_batch, observation_num):\n",
    "        \"\"\"Trains networks to fit given parameters\"\"\"\n",
    "\n",
    "        # nothing has changed from the original implementation, except for changing the input dimension 'reshape'\n",
    "        batch_size = s_batch.shape[0]\n",
    "        target = np.zeros((batch_size, 1))\n",
    "        new_proba = np.zeros((batch_size, NUM_ACTIONS))\n",
    "        last_action=np.zeros((batch_size, NUM_ACTIONS))\n",
    "        \n",
    "        #training of the action state value networks\n",
    "        last_action=np.zeros((batch_size, NUM_ACTIONS))\n",
    "\n",
    "        for i in range(batch_size):\n",
    "            last_action[i,a_batch[i]] = 1\n",
    "\n",
    "            v_t = self.model_value_target.predict(s_batch[i].reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "            \n",
    "            self.qvalue_evolution.append(v_t[0])\n",
    "            fut_action = self.model_value_target.predict(s2_batch[i].reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "\n",
    "            target[i,0] = r_batch[i] + (1 - d_batch[i]) * DECAY_RATE * fut_action   \n",
    "\n",
    "                \n",
    "        loss = self.model_Q.train_on_batch([s_batch, last_action], target)\n",
    "        loss_2 = self.model_Q2.train_on_batch([s_batch, last_action], target) \n",
    "            \n",
    "        #training of the policy\n",
    "        \n",
    "        for i in range(batch_size):\n",
    "            self.life_spent += 1\n",
    "            temp =  1 / np.log(self.life_spent) / 2\n",
    "            new_values = self.model_Q.predict([np.tile(s_batch[i].reshape(1, self.observation_size),(NUM_ACTIONS,1)),\n",
    "                                                      np.eye(NUM_ACTIONS)]).reshape(1,-1)\n",
    "            new_values -= np.amax(new_values, axis=-1)\n",
    "            new_proba[i] = np.exp(new_values / temp) / np.sum(np.exp(new_values / temp), axis=-1)\n",
    "            \n",
    "        loss_policy = self.model_policy.train_on_batch(s_batch, new_proba)\n",
    "    \n",
    "        #training of the value_function\n",
    "        value_target = np.zeros(batch_size)\n",
    "        for i in range(batch_size):\n",
    "            target_pi = self.model_policy.predict(s_batch[i].reshape(1, self.observation_size*NUM_FRAMES), batch_size = 1)\n",
    "            action_v1 = self.model_Q.predict([np.tile(s_batch[i].reshape(1, self.observation_size),(NUM_ACTIONS,1)),\n",
    "                                                      np.eye(NUM_ACTIONS)]).reshape(1,-1)\n",
    "            action_v2 = self.model_Q2.predict([np.tile(s_batch[i].reshape(1, self.observation_size),(NUM_ACTIONS,1)),\n",
    "                                                      np.eye(NUM_ACTIONS)]).reshape(1,-1)\n",
    "            value_target[i] = np.fmin(action_v1[0,a_batch[i]], action_v2[0,a_batch[i]]) - np.sum(target_pi * np.log(target_pi + 1e-6))\n",
    "        \n",
    "        loss_value = self.model_value.train_on_batch(s_batch, value_target.reshape(-1,1))\n",
    "        \n",
    "        self.Is_nan = np.isnan(loss) + np.isnan(loss_2) + np.isnan(loss_policy) + np.isnan(loss_value)\n",
    "        # Print the loss every 100 iterations.\n",
    "        if observation_num % 100 == 0:\n",
    "            print(\"We had a loss equal to \", loss, loss_2, loss_policy, loss_value)\n",
    "\n",
    "    def save_network(self, path):\n",
    "        # Saves model at specified path as h5 file\n",
    "        # nothing has changed\n",
    "        self.model_policy.save(\"policy_\"+path)\n",
    "        self.model_value_target.save(\"value_\"+path)\n",
    "        print(\"Successfully saved network.\")\n",
    "\n",
    "    def load_network(self, path):\n",
    "        # nothing has changed\n",
    "        self.model_policy = load_model(\"policy_\"+path)\n",
    "        elf.model_value_target = load_model(\"value_\"+path)\n",
    "        print(\"Succesfully loaded network.\")\n",
    "\n",
    "    def target_train(self):\n",
    "        # nothing has changed from the original implementation\n",
    "        model_weights = self.model_value.get_weights() \n",
    "        target_model_weights = self.model_value_target.get_weights()\n",
    "        for i in range(len(model_weights)):\n",
    "            target_model_weights[i] = TAU * model_weights[i] + (1 - TAU) * target_model_weights[i]\n",
    "        self.model_value_target.set_weights(model_weights)\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We first show how the agent code looks like without the \"utilities\" to train it. As we can see, defining this agent is pretty simple. The only real method that need to be adapted it \"`my_act`\" that is a method of 2 lines of code."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Credit Abhinav Sagar: \n",
    "# https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial\n",
    "# Code under MIT license, available at:\n",
    "# https://github.com/abhinavsagar/Reinforcement-Learning-Tutorial/blob/master/LICENSE\n",
    "\n",
    "from grid2op.Parameters import Parameters\n",
    "from grid2op.Agent import AgentWithConverter\n",
    "from grid2op.Converter import IdToAct\n",
    "import pdb\n",
    "\n",
    "\n",
    "class DeepQAgent(AgentWithConverter):\n",
    "    # first change: An Agent must derived from grid2op.Agent (in this case MLAgent, because we manipulate vector instead\n",
    "    # of classes)\n",
    "    \n",
    "    def convert_obs(self, observation):\n",
    "        return observation.to_vect()\n",
    "        \n",
    "    def my_act(self, transformed_observation, reward, done=False):\n",
    "        if self.deep_q is None:\n",
    "            self.init_deep_q(transformed_observation)\n",
    "        predict_movement_int, *_ = self.deep_q.predict_movement(transformed_observation, epsilon=0.0)\n",
    "        # print(\"predict_movement_int: {}\".format(predict_movement_int))\n",
    "        return predict_movement_int\n",
    "    \n",
    "    def init_deep_q(self, transformed_observation):\n",
    "        if self.deep_q is None:\n",
    "            # the first time an observation is observed, I set up the neural network with the proper dimensions.\n",
    "            if self.mode == \"DQN\":\n",
    "                cls = DeepQ\n",
    "            elif self.mode == \"DDQN\":\n",
    "                cls = DuelQ\n",
    "            elif self.mode == \"RealQ\":\n",
    "                cls = RealQ\n",
    "            elif self.mode == \"SAC\":\n",
    "                cls = SAC\n",
    "            else:\n",
    "                raise RuntimeError(\"Unknown neural network named \\\"{}\\\"\".format(self.mode))\n",
    "            self.deep_q = cls(self.action_space.size(), observation_size=transformed_observation.shape[0], lr=self.lr)\n",
    "            \n",
    "    def __init__(self, action_space, mode=\"DDQN\", lr=1e-5):\n",
    "        # this function has been adapted.\n",
    "        \n",
    "        # to built a AgentWithConverter, we need an action_space. \n",
    "        # No problem, we add it in the constructor.\n",
    "        AgentWithConverter.__init__(self, action_space, action_space_converter=IdToAct)\n",
    "        \n",
    "        # and now back to the origin implementation\n",
    "        self.replay_buffer = ReplayBuffer(BUFFER_SIZE)\n",
    "        \n",
    "        # compare to original implementation, i don't know the observation space size. \n",
    "        # Because it depends on the component of the observation we want to look at. So these neural network will\n",
    "        # be initialized the first time an observation is observe.\n",
    "        self.deep_q = None\n",
    "        self.mode = mode\n",
    "        self.lr=lr\n",
    "    \n",
    "    def load_network(self, path):\n",
    "        # not modified compare to original implementation\n",
    "        self.deep_q.load_network(path)\n",
    "    \n",
    "    def convert_process_buffer(self):\n",
    "        \"\"\"Converts the list of NUM_FRAMES images in the process buffer\n",
    "        into one training sample\"\"\"\n",
    "        # here i simply concatenate the action in case of multiple action in the \"buffer\"\n",
    "        # this function existed in the original implementation, bus has been adapted.\n",
    "        return np.concatenate(self.process_buffer)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And now we will also define some utility class (as defined in the blog post) to make the training easier."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "from grid2op.Reward import L2RPNReward\n",
    "from grid2op.Reward import RedispReward\n",
    "\n",
    "class L2RPNReward_LoadWise(L2RPNReward):\n",
    "    \"\"\"\n",
    "    Update the L2RPN reward to take into account the fact that a change in the loads sum shall not be allocated as reward for the agent.\n",
    "    \n",
    "    \"\"\"\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "    \n",
    "    def initialize(self, env):\n",
    "        super().initialize(env)\n",
    "        self.reward_min = - 10 * env.backend.n_line\n",
    "        self.previous_loads = self.reward_max * np.ones(env.backend.n_line)\n",
    "        \n",
    "    def __call__(self, action, env, has_error, is_done, is_illegal, is_ambiguous):\n",
    "        if not is_done and not has_error:\n",
    "            line_cap = self._L2RPNReward__get_lines_capacity_usage(env)\n",
    "            \n",
    "            new_loads, _, _ = env.backend.loads_info()\n",
    "            new_flows = np.abs(env.backend.get_line_flow())\n",
    "            loads_variation = (np.sum(new_loads) - np.sum(self.previous_loads)) / np.sum(self.previous_loads)\n",
    "            \n",
    "            res = np.sum(line_cap  + loads_variation)\n",
    "        else:\n",
    "            # no more data to consider, no powerflow has been run, reward is what it is\n",
    "            res = self.reward_min\n",
    "        return res\n",
    "      \n",
    "class L2RPNReward_LoadWise_ActionWise(L2RPNReward):\n",
    "    \"\"\"\n",
    "    Update the L2RPN reward to take into account the fact that a change in the loads sum shall not be allocated as reward for the agent.\n",
    "    \n",
    "    \"\"\"\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        \n",
    "    def initialize(self, env):\n",
    "        super().initialize(env)\n",
    "        self.reward_min = - 10 * env.backend.n_line\n",
    "        self.previous_loads = self.reward_max * np.ones(env.backend.n_line)\n",
    "        self.last_action = env.helper_action_env({})\n",
    "        \n",
    "    def __call__(self, action, env, has_error, is_done, is_illegal, is_ambiguous):\n",
    "        if not is_done and not has_error:\n",
    "            line_cap = self._L2RPNReward__get_lines_capacity_usage(env)\n",
    "            \n",
    "            new_loads, _, _ = env.backend.loads_info()\n",
    "            new_flows = np.abs(env.backend.get_line_flow())\n",
    "            loads_variation = (np.sum(new_loads) - np.sum(self.previous_loads)) / np.sum(self.previous_loads)\n",
    "            \n",
    "            res = np.sum(line_cap  + loads_variation)\n",
    "        else:\n",
    "            # no more data to consider, no powerflow has been run, reward is what it is\n",
    "            res = self.reward_min\n",
    "            \n",
    "        res -= (action != env.helper_action_env({})) * (action == self.last_action) * env.backend.n_line / 2\n",
    "            \n",
    "        self.last_action = action\n",
    "        \n",
    "        return res\n",
    "      \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "class TrainAgent(object):\n",
    "    def __init__(self, agent, reward_fun=RedispReward, env=None):\n",
    "        self.agent = agent\n",
    "        self.reward_fun = reward_fun\n",
    "        self.env = env\n",
    "        \n",
    "    def _build_valid_env(self):\n",
    "        # now we are creating a valid Environment\n",
    "        # it's mandatory because no environment are created when the agent is \n",
    "        # an Agent should not get direct access to the environment, but can interact with it only by:\n",
    "        # * receiving reward\n",
    "        # * receiving observation\n",
    "        # * sending action\n",
    "        \n",
    "        close_env = False\n",
    "        \n",
    "        if self.env is None:\n",
    "            self.env = grid2op.make(action_class=type(self.agent.action_space({})),\n",
    "                                    reward_class=self.reward_fun)\n",
    "            close_env = True\n",
    "                               \n",
    "        # I make sure the action space of the user and the environment are the same.\n",
    "        if not isinstance(self.agent.init_action_space, type(self.env.action_space)):\n",
    "            raise RuntimeError(\"Imposssible to build an agent with 2 different action space\")\n",
    "        if not isinstance(self.env.action_space, type(self.agent.init_action_space)):\n",
    "            raise RuntimeError(\"Imposssible to build an agent with 2 different action space\")\n",
    "        \n",
    "        # A buffer that keeps the last `NUM_FRAMES` images\n",
    "        self.agent.replay_buffer.clear()\n",
    "        self.agent.process_buffer = []\n",
    "        \n",
    "        # make sure the environment is reset\n",
    "        obs = self.env.reset()\n",
    "        self.agent.process_buffer.append(self.agent.convert_obs(obs))\n",
    "        do_nothing = self.env.action_space()\n",
    "        for _ in range(NUM_FRAMES-1):\n",
    "            # Initialize buffer with the first frames\n",
    "            s1, r1, _, _ = self.env.step(do_nothing)\n",
    "            self.agent.process_buffer.append(self.agent.convert_obs(s1))            \n",
    "        return close_env\n",
    "    \n",
    "    def train(self, num_frames, env=None):\n",
    "        # this function existed in the original implementation, but has been slightly adapted.\n",
    "        \n",
    "        # first we create an environment or make sure the given environment is valid\n",
    "        close_env = self._build_valid_env()\n",
    "        \n",
    "        # bellow that, only slight modification has been made. They are highlighted\n",
    "        observation_num = 0\n",
    "        curr_state = self.agent.convert_process_buffer()\n",
    "        epsilon = INITIAL_EPSILON\n",
    "        alive_frame = 0\n",
    "        total_reward = 0\n",
    "\n",
    "        while observation_num < num_frames:\n",
    "            if observation_num % 1000 == 999:\n",
    "                print((\"Executing loop %d\" %observation_num))\n",
    "\n",
    "            # Slowly decay the learning rate\n",
    "            if epsilon > FINAL_EPSILON:\n",
    "                epsilon -= (INITIAL_EPSILON-FINAL_EPSILON)/EPSILON_DECAY\n",
    "\n",
    "            initial_state = self.agent.convert_process_buffer()\n",
    "            self.agent.process_buffer = []\n",
    "\n",
    "            # it's a bit less convenient that using the SpaceInvader environment.\n",
    "            # first we need to initiliaze the neural network\n",
    "            self.agent.init_deep_q(curr_state)\n",
    "            # then we need to predict the next move\n",
    "            predict_movement_int, predict_q_value = self.agent.deep_q.predict_movement(curr_state, epsilon)\n",
    "            # and then we convert it to a valid action\n",
    "            act = self.agent.convert_act(predict_movement_int)\n",
    "            \n",
    "            reward, done = 0, False\n",
    "            for i in range(NUM_FRAMES):\n",
    "                temp_observation_obj, temp_reward, temp_done, _ = self.env.step(act)\n",
    "                # here it has been adapted too. The observation get from the environment is\n",
    "                # first converted to vector\n",
    "                \n",
    "                # below this line no changed have been made to the original implementation.\n",
    "                reward += temp_reward\n",
    "                self.agent.process_buffer.append(self.agent.convert_obs(temp_observation_obj))\n",
    "                done = done | temp_done\n",
    "\n",
    "            if done:\n",
    "                print(\"Lived with maximum time \", alive_frame)\n",
    "                print(\"Earned a total of reward equal to \", total_reward)\n",
    "                # reset the environment\n",
    "                self.env.reset()\n",
    "                \n",
    "                alive_frame = 0\n",
    "                total_reward = 0\n",
    "\n",
    "            new_state = self.agent.convert_process_buffer()\n",
    "            self.agent.replay_buffer.add(initial_state, predict_movement_int, reward, done, new_state)\n",
    "            total_reward += reward\n",
    "            if self.agent.replay_buffer.size() > MIN_OBSERVATION:\n",
    "                s_batch, a_batch, r_batch, d_batch, s2_batch = self.agent.replay_buffer.sample(MINIBATCH_SIZE)\n",
    "                self.agent.deep_q.train(s_batch, a_batch, r_batch, d_batch, s2_batch, observation_num)\n",
    "                self.agent.deep_q.target_train()\n",
    "\n",
    "            # Save the network every 1000 iterations after 50000 iterations\n",
    "            if observation_num > 50000 and observation_num % 1000 == 0 and self.agent.deep_q.Is_nan == 0:\n",
    "                print(\"Saving Network\")\n",
    "                self.agent.deep_q.save_network(\"saved_agent_\"+self.agent.mode+\".h5\")\n",
    "\n",
    "            alive_frame += 1\n",
    "            observation_num += 1\n",
    "            \n",
    "        if close_env:\n",
    "            print(\"closing env\")\n",
    "            self.env.close()\n",
    "            \n",
    "    def calculate_mean(self, num_episode = 100, env=None):\n",
    "        # this method has been only slightly adapted from the original implementation\n",
    "        \n",
    "        # Note that it is NOT the recommended method to evaluate an Agent. Please use \"Grid2Op.Runner\" instead\n",
    "        \n",
    "        # first we create an environment or make sure the given environment is valid\n",
    "        close_env = self._build_valid_env()\n",
    "        \n",
    "        reward_list = []\n",
    "        print(\"Printing scores of each trial\")\n",
    "        for i in range(num_episode):\n",
    "            done = False\n",
    "            tot_award = 0\n",
    "            self.env.reset()\n",
    "            while not done:\n",
    "                state = self.convert_process_buffer()\n",
    "                \n",
    "                # same adapation as in \"train\" function. \n",
    "                predict_movement_int = self.agent.deep_q.predict_movement(state, 0.0)[0]\n",
    "                predict_movement = self.agent.convert_act(predict_movement_int)\n",
    "                \n",
    "                # same adapation as in the \"train\" funciton\n",
    "                observation_obj, reward, done, _ = self.env.step(predict_movement)\n",
    "                observation_vect_full = observation_obj.to_vect()\n",
    "                \n",
    "                tot_award += reward\n",
    "                self.process_buffer.append(observation)\n",
    "                self.process_buffer = self.process_buffer[1:]\n",
    "            print(tot_award)\n",
    "            reward_list.append(tot_award)\n",
    "            \n",
    "        if close_env:\n",
    "            self.env.close()\n",
    "        return np.mean(reward_list), np.std(reward_list)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### b) Training the model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we can define the model (agent), and then train it.\n",
    "\n",
    "This is done exactly the same way as in the Abhinav Sagar implementation.\n",
    "\n",
    "**NB** The code bellow can take a few minutes to run. It's training a Deep Reinforcement Learning Agent afterall. It this takes too long on your machine, you can always decrease the \"nb_frame\", and set it to 1000 for example. In this case, the Agent will probably not be really good.\n",
    "\n",
    "**NB** For a real Agent, it would take much longer to train."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/benjamin/Documents/grid2op_test/getting_started/grid2op/MakeEnv.py:668: UserWarning:\n",
      "\n",
      "Your are using only 2 chronics for this environment. More can be download by running, from a command line:\n",
      "python -m grid2op.download --name \"case14_redisp\" --path_save PATH\\WHERE\\YOU\\WANT\\TO\\DOWNLOAD\\DATA\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully constructed networks.\n",
      "Lived with maximum time  87\n",
      "Earned a total of reward equal to  99318.90852576918\n",
      "Lived with maximum time  8\n",
      "Earned a total of reward equal to  7582.116955233524\n",
      "Successfully saved network.\n"
     ]
    }
   ],
   "source": [
    "nb_frame = 100\n",
    "\n",
    "env = make(name_env=\"case14_redisp\")\n",
    "my_agent = DeepQAgent(env.action_space, mode=\"DDQN\")\n",
    "trainer = TrainAgent(agent=my_agent, env=env, reward_fun=RedispReward)\n",
    "trainer.train(nb_frame)\n",
    "trainer.agent.deep_q.save_network(\"saved_agent_\"+trainer.agent.mode+\".h5\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# To see whether the agent learnt something during training: lets plot the learning curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.0, 100.0)"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x1440 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "\n",
    "plt.figure(figsize=(30,20))\n",
    "plt.plot(my_agent.deep_q.qvalue_evolution)\n",
    "plt.axhline(y=0, linewidth=3, color='red')\n",
    "plt.xlim(0, len(my_agent.deep_q.qvalue_evolution))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The learning curve shown above is really poor. It's because the agent has not been trained for a long time (and because it uses a very poor input data). If you train this agent for approximately 10-12 hours, using only the relative flow (`obs.rho` see the last section of this notebook for an example) you will get the following:\n",
    "\n",
    "![](img/trained_agent.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## III) Evaluating the Agent"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And now, time to test this trained agent.\n",
    "\n",
    "To do that, we have multiple choices.\n",
    "\n",
    "Either we recode the \"DeepQAgent\" class to load the stored weights (that have been saved during trainig) when it is initialized (not covered in this notebook), or we can also directly specify the \"instance\" of the Agent to use in the Grid2Op Runner.\n",
    "\n",
    "To do that, it's fairly simple. First, you need to specify that you won't use the \"*agentClass*\" argument, by setting it to ``None``, and secondly you simply provide the agent to use in the *agentInstance* argument.\n",
    "\n",
    "**NB** If you don't do that, the Runner will be created (the constructor will raise an exception). And if you choose to use the \"*agentClass*\" argument, your agent will be reloaded from scratch. So **if it doesn't load the weights** it will behave as a non trained agent, unlikely to perform well on the task."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### III.A) Evaluate the Agent"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now that we have \"successfully\" trained our Agent, we will evaluating it. As opposed to the trainining, the evaluation is done classically using a standard Runner.\n",
    "\n",
    "Note that the Runner will use a \"scoring function\" that might be different from the \"reward function\" used during training. In our case, it's not. We use the `L2RPNReward` in both cases.\n",
    "\n",
    "In the code bellow, we commented on what can be different and what must be identical for training and evaluation of model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "from grid2op.Runner import Runner\n",
    "from grid2op.Chronics import GridStateFromFileWithForecasts, Multifolder\n",
    "scoring_function = L2RPNReward\n",
    "\n",
    "dict_params = env.get_params_for_runner()\n",
    "dict_params[\"gridStateclass_kwargs\"][\"max_iter\"] =  max_iter\n",
    "# make a runner from an intialized environment\n",
    "runner = Runner(**dict_params)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Run the Agent and save the results. As opposed to the multiple times we exposed the \"runner.run\" call, we never really dive into the \"path_save\" argument. This path allows you to save lots of information about your Agent behaviour. Please All the informations present are shown on the documentation [here](https://grid2op.readthedocs.io/en/latest/runner.html)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The results for the trained agent are:\n",
      "\tFor chronics located at 0\n",
      "\t\t - cumulative reward: 121368.512117\n",
      "\t\t - number of time steps completed: 100 / 100\n"
     ]
    }
   ],
   "source": [
    "import shutil\n",
    "path_save=\"trained_agent_log\"\n",
    "\n",
    "# delete the previous stored results\n",
    "if os.path.exists(path_save):\n",
    "    shutil.rmtree(path_save)\n",
    "\n",
    "# run the episode\n",
    "res = runner.run(nb_episode=1, path_save=path_save)\n",
    "print(\"The results for the trained agent are:\")\n",
    "for _, chron_name, cum_reward, nb_time_step, max_ts in res:\n",
    "    msg_tmp = \"\\tFor chronics located at {}\\n\".format(chron_name)\n",
    "    msg_tmp += \"\\t\\t - cumulative reward: {:.6f}\\n\".format(cum_reward)\n",
    "    msg_tmp += \"\\t\\t - number of time steps completed: {:.0f} / {:.0f}\".format(nb_time_step, max_ts)\n",
    "    print(msg_tmp)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### III.B) Inspect the Agent "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Please refer to the official document for more information about the content of the directory where the data are saved. Note that the saving of the information is triggered by the \"path_save\" argument sent to the \"runner.run\" function.\n",
    "\n",
    "Some information that will be present in this repository are:\n",
    "If enabled, the :class:`Runner` will save the information in a structured way. For each episode there will be a folder\n",
    "with:\n",
    "\n",
    "  - \"episode_meta.json\" that represents some meta information about:\n",
    "\n",
    "    - \"backend_type\": the name of the `grid2op.Backend` class used\n",
    "    - \"chronics_max_timestep\": the **maximum** number of timestep for the chronics used\n",
    "    - \"chronics_path\": the path where the temporal data (chronics) are located\n",
    "    - \"env_type\": the name of the `grid2op.Environment` class used.\n",
    "    - \"grid_path\": the path where the powergrid has been loaded from\n",
    "\n",
    "  - \"episode_times.json\": gives some information about the total time spend in multiple part of the runner, mainly the\n",
    "    `grid2op.Agent` (and especially its method `grid2op.Agent.act`) and amount of time spent in the\n",
    "    `grid2op.Environment`\n",
    "\n",
    "  - \"_parameters.json\": is a representation as json of a the `grid2op.Parameters` used for this episode\n",
    "  - \"rewards.npy\" is a numpy 1d array giving the rewards at each time step. We adopted the convention that the stored\n",
    "    reward at index `i` is the one observed by the agent at time `i` and **NOT** the reward sent by the\n",
    "    `grid2op.Environment` after the action has been implemented.\n",
    "  - \"exec_times.npy\" is a numpy 1d array giving the execution time of each time step of the episode\n",
    "  - \"actions.npy\" gives the actions that has been taken by the `grid2op.Agent`. At row `i` of \"actions.npy\" is a\n",
    "    vectorized representation of the action performed by the agent at timestep `i` *ie.* **after** having observed\n",
    "    the observation present at row `i` of \"observation.npy\" and the reward showed in row `i` of \"rewards.npy\".\n",
    "  - \"disc_lines.npy\" gives which lines have been disconnected during the simulation of the cascading failure at each\n",
    "    time step. The same convention as for \"rewards.npy\" has been adopted. This means that the powerlines are\n",
    "    disconnected when the `grid2op.Agent` takes the `grid2op.Action` at time step `i`.\n",
    "  - \"observations.npy\" is a numpy 2d array reprensenting the `grid2op.Observation` at the disposal of the\n",
    "    `grid2op.Agent` when he took his action.\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can first look at the repository were the data are stored:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['dict_env_modification_space.json',\n",
       " 'dict_action_space.json',\n",
       " 'dict_observation_space.json',\n",
       " '0']"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import os\n",
    "os.listdir(path_save)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we can see, there is only one folder there. It's named \"1\" because, in the original data, this came from the folder named \"1\" (the original data are located at \"/home/donnotben/.local/lib/python3.6/site-packages/grid2op/data/test_multi_chronics/\")\n",
    "\n",
    "If there were multiple episode, each episode would have it's own folder, with a name as resemblant as possible to the origin name of the data. This is done to ease the studying of the results.\n",
    "\n",
    "Now let's see what is inside this folder:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['rewards.npy',\n",
       " 'episode_meta.json',\n",
       " 'disc_lines_cascading_failure.npy',\n",
       " 'agent_exec_times.npy',\n",
       " 'actions.npy',\n",
       " 'episode_times.json',\n",
       " 'observations.npy',\n",
       " '_parameters.json',\n",
       " 'other_rewards.json',\n",
       " 'env_modifications.npy']"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "os.listdir(os.path.join(path_save,\"0\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can for example load the \"actions\" performed by the Agent, and have a look at them.\n",
    "\n",
    "To do that we will load the action array (represented as vector) and use the action_space to convert it back into valid action class."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_actions = np.load(os.path.join(\"trained_agent_log\", \"0\", \"actions.npy\"))\n",
    "li_actions = []\n",
    "for i in range(all_actions.shape[0]):\n",
    "    try:\n",
    "        tmp = runner.env.action_space.from_vect(all_actions[i,:])\n",
    "        li_actions.append(tmp)\n",
    "    except:\n",
    "        break"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This allows us to have a deeper look at the action, and their effect. Note that here, we used action that can only **set** the line status, so looking at their effect is pretty straightforward.\n",
    "\n",
    "Also, note that as oppose to \"change\", if a powerline is already connected, trying to **set** it as connected has absolutely no impact."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "line_disc = 0\n",
    "line_reco = 0\n",
    "for act in li_actions:\n",
    "    dict_ = act.as_dict()\n",
    "    if \"set_line_status\" in dict_:\n",
    "        line_reco +=  dict_[\"set_line_status\"][\"nb_connected\"]\n",
    "        line_disc +=  dict_[\"set_line_status\"][\"nb_disconnected\"]\n",
    "line_reco"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As wa can see for our event, the agent always try to reconnect a powerline. As all lines are alway reconnected, this Agent does basically nothing."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can also do the same kind of post analysis for the observation, even though here, as the observations come from files, it's probably not particularly intersting."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_observations = np.load(os.path.join(\"trained_agent_log\", \"0\", \"observations.npy\"))\n",
    "li_observations = []\n",
    "nb_real_disc = 0\n",
    "for i in range(all_observations.shape[0]):\n",
    "    try:\n",
    "        tmp = runner.env.observation_space.from_vect(all_observations[i,:])\n",
    "        li_observations.append(tmp)\n",
    "        nb_real_disc += (np.sum(tmp.line_status) - tmp.line_status.shape[0])\n",
    "    except:\n",
    "        break\n",
    "nb_real_disc"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can also look at the type of action the agent did:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of different actions played: 1\n"
     ]
    }
   ],
   "source": [
    "actions_count = {}\n",
    "for act in li_actions:\n",
    "    act_as_vect = tuple(act.to_vect())\n",
    "    if not act_as_vect in actions_count:\n",
    "        actions_count[act_as_vect] = 0\n",
    "    actions_count[act_as_vect] += 1\n",
    "print(\"Number of different actions played: {}\".format(len(actions_count)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "This action will:\n",
      "\t - NOT change anything to the injections\n",
      "\t - NOT perform any redispatching action\n",
      "\t - NOT force any line status\n",
      "\t - NOT switch any line status\n",
      "\t - NOT switch anything in the topology\n",
      "\t - NOT force any particular bus configuration\n"
     ]
    }
   ],
   "source": [
    "print(runner.env.action_space.from_vect(np.array(list(actions_count.keys())[0])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The agent only did one action (note that this number can really vary on the number of training step and the . This is not really good, the agent didn't learn anything."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## IV) Improve your Agent "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we saw, the agent we develop was not really interesting. To improve it, we could think about:\n",
    "\n",
    "- a better encoding of the observation. For now everything is fed to the neural network, without any normalization of any kind. This is a real problem for learning algorithm.\n",
    "- a better neural network architecture (as said, we didn't pay any attention to it in our model)\n",
    "- train it for a longer time\n",
    "- adapt the learning rate and all the meta parameters of the learning algorithm.\n",
    "- etc.\n",
    "\n",
    "In this notebook, we will focus on changing the observation representation, by only feeding the agent only some informations.\n",
    "\n",
    "To do so, the only modification we need to do is to modify the way the observation are converted. So the \"*convert_obs*\" method, and that is it. Nothing else need to be changed. Here for example we could think of only using the flow ratio (*i.e.,* the current flows divided by the thermal limits) as part of the observation (named rho) instead of feeding the whole observation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "class DeepQAgent_Improved(DeepQAgent):\n",
    "    def __init__(self, action_space, mode=\"DDQN\"):\n",
    "        DeepQAgent.__init__(self, action_space, mode=mode)\n",
    "    \n",
    "    def convert_obs(self, observation):\n",
    "        return observation.rho"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And we can reuse exactly the same code to train it :-)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully constructed networks.\n",
      "Lived with maximum time  56\n",
      "Earned a total of reward equal to  3232.3359374684956\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x1440 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "my_agent = DeepQAgent_Improved(env.action_space, mode=\"DDQN\")\n",
    "trainer = TrainAgent(agent=my_agent, env=env)\n",
    "trainer.train(nb_frame)\n",
    "\n",
    "plt.figure(figsize=(30,20))\n",
    "plt.plot(my_agent.deep_q.qvalue_evolution)\n",
    "plt.axhline(y=0, linewidth=3, color='red')\n",
    "_ = plt.xlim(0, len(my_agent.deep_q.qvalue_evolution))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}