{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import logging\n",
    "logging.basicConfig(level=logging.DEBUG)\n",
    "from sklearn.metrics import accuracy_score\n",
    "from sklearn.svm import SVC\n",
    "\n",
    "from wtk.utilities import get_ucr_dataset, krein_svm_grid_search\n",
    "from wtk import transform_to_dist_matrix, get_kernel_matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Read UCR data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, y_train, X_test, y_test = get_ucr_dataset('../data/UCR/raw_data/', 'ItalyPowerDemand')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Compute wasserstein distance matrices with subsequent length $k=10$\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "D_train, D_test = transform_to_dist_matrix(X_train, X_test, 10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Run the grid search"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Starting analysis\n",
      "INFO:root:Accuracy = 95.34\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best C: 1.0\n",
      "Best gamma: 0.1\n"
     ]
    }
   ],
   "source": [
    "svm_clf = krein_svm_grid_search(D_train, D_test, y_train, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Alternatively: Get the kernel matrices computed from the distance matrices ..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "K_train = get_kernel_matrix(D_train, psd=True, gamma=0.2)\n",
    "K_test = get_kernel_matrix(D_test, psd=False, gamma=0.2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### ... and train your own classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9640427599611273"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "clf = SVC(C=5, kernel='precomputed')\n",
    "clf.fit(K_train, y_train)\n",
    "y_pred = clf.predict(K_test)\n",
    "accuracy_score(y_test, y_pred)"
   ]
  }
 ],
 "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": "3.7.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}