{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Uniform Manifold Approximation and Projection (UMAP)\n",
    "UMAP is a dimensionality reduction algorithm which performs non-linear dimension reduction. It can also be used for visualization of the dataset. The UMAP model implemented in cuml allows the user to set the following parameter values:\n",
    "1.\tn_neighbors: number of neighboring samples used for manifold approximation. Larger values result in more global views of the manifold, while smaller values result in more local data being preserved (default = 15)\n",
    "2.\tn_components: the dimension of the space to embed into (default = 2)\n",
    "3.\tn_epochs: number of training epochs to be used in optimizing the low dimensional embedding (default = None)\n",
    "4.\tlearning_rate: initial learning rate for the embedding optimization (default = 1.0)\n",
    "5.\tinit: the low dimensional embedding to use: a. 'spectral': use a spectral embedding of the fuzzy 1-skeleton b. 'random': assign initial embedding positions at random (default = 'spectral')\n",
    "6.\tmin_dist: the minimum distance that should be present between embedded points (default = 0.1)\n",
    "7.\tspread: determines how clustered the embedded points will be (default = 1.0)\n",
    "8.\tset_op_mix_ratio: is the ratio of pure fuzzy union to intersection. If the value is 1.0 then it will be a pure fuzzy union and for the value of 0.0 it will be a pure fuzzy interpolation (default = 1.0)\n",
    "9.\tlocal_connectivity: number of nearest neighbors that should be assumed to be connected at a local level. It should be not more than the local intrinsic dimension of the manifold (default = 1)\n",
    "10.\trepulsion_strength: weighting applied to negative samples in low dimensional embedding optimization. Values > 1 implements a higher negative value to the samples (default = 1.0)\n",
    "11.\tnegative_sample_rate: the rate at which the negative samples should be selected per positive sample during the optimization process (default = 5)\n",
    "12.\ttransform_queue_size: embedding new points using a trained model_ will control how aggressively to search for nearest neighbors (default = 4.0)\n",
    "13.\tverbose: bool (default False)\n",
    "\n",
    "The cuml implemetation of the UMAP model has the following functions that one can run:\n",
    "1.\tfit: it fits the dataset into an embedded space\n",
    "2.\tfit_transform: it fits the dataset into an embedded space and returns the transformed output\n",
    "3.\ttransform: it transforms the dataset into an existing embedded space and returns the low dimensional output\n",
    "\n",
    "The model can take array-like objects, either in host as NumPy arrays or in device (as Numba or _cuda_array_interface_compliant), as well  as cuDF DataFrames. In order to convert your dataset to cudf format please read the cudf documentation on https://rapidsai.github.io/projects/cudf/en/latest/. For additional information on the UMAP model please refer to the documentation on https://rapidsai.github.io/projects/cuml/en/0.6.0/api.html#cuml.UMAP"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import cudf\n",
    "import os\n",
    "\n",
    "from sklearn import datasets\n",
    "from sklearn.metrics import adjusted_rand_score\n",
    "from sklearn.cluster import KMeans\n",
    "from sklearn.manifold.t_sne import trustworthiness\n",
    "\n",
    "from cuml.manifold.umap import UMAP"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Running cuml's UMAP model on blobs dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create a blobs dataset with 500 samples and 10 features each\n",
    "data, labels = datasets.make_blobs(\n",
    "    n_samples=500, n_features=10, centers=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# using the cuml UMAP algorithm to reduce the features of the dataset and store\n",
    "embedding = UMAP().fit_transform(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.0\n"
     ]
    }
   ],
   "source": [
    "# calculate the score of the results obtained using cuml's algorithm and sklearn kmeans\n",
    "score = adjusted_rand_score(labels,\n",
    "            KMeans(5).fit_predict(embedding))\n",
    "print(score) # should equal 1.0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Running cuml's UMAP model on iris dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load the iris dataset from sklearn and extract the required information\n",
    "iris = datasets.load_iris()\n",
    "data = iris.data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# define the cuml UMAP model and use fit_transform function to obtain the low dimensional output of the input dataset\n",
    "embedding = UMAP(\n",
    "    n_neighbors=10, min_dist=0.01,  init=\"random\"\n",
    ").fit_transform(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9741363073110285\n"
     ]
    }
   ],
   "source": [
    "# calculate the trust worthiness of the results obtaind from the cuml UMAP\n",
    "trust = trustworthiness(iris.data, embedding, 10)\n",
    "print(trust)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[5.1 3.5 1.4 0.2]\n",
      " [4.9 3.  1.4 0.2]\n",
      " [4.6 3.1 1.5 0.2]\n",
      " [5.  3.6 1.4 0.2]\n",
      " [4.6 3.4 1.4 0.3]\n",
      " [5.  3.4 1.5 0.2]\n",
      " [4.4 2.9 1.4 0.2]\n",
      " [4.9 3.1 1.5 0.1]\n",
      " [5.4 3.7 1.5 0.2]\n",
      " [4.8 3.4 1.6 0.2]\n",
      " [4.8 3.  1.4 0.1]\n",
      " [4.3 3.  1.1 0.1]\n",
      " [5.7 4.4 1.5 0.4]\n",
      " [5.4 3.9 1.3 0.4]\n",
      " [5.1 3.5 1.4 0.3]\n",
      " [5.7 3.8 1.7 0.3]\n",
      " [5.1 3.8 1.5 0.3]\n",
      " [5.4 3.4 1.7 0.2]\n",
      " [5.1 3.3 1.7 0.5]\n",
      " [4.8 3.4 1.9 0.2]\n",
      " [5.  3.  1.6 0.2]\n",
      " [5.  3.4 1.6 0.4]\n",
      " [5.2 3.4 1.4 0.2]\n",
      " [4.7 3.2 1.6 0.2]\n",
      " [4.8 3.1 1.6 0.2]\n",
      " [5.2 4.1 1.5 0.1]\n",
      " [5.5 4.2 1.4 0.2]\n",
      " [5.  3.2 1.2 0.2]\n",
      " [4.9 3.6 1.4 0.1]\n",
      " [4.4 3.  1.3 0.2]\n",
      " [5.  3.5 1.3 0.3]\n",
      " [4.5 2.3 1.3 0.3]\n",
      " [4.4 3.2 1.3 0.2]\n",
      " [5.  3.5 1.6 0.6]\n",
      " [5.1 3.8 1.9 0.4]\n",
      " [4.8 3.  1.4 0.3]\n",
      " [5.1 3.8 1.6 0.2]\n",
      " [4.6 3.2 1.4 0.2]\n",
      " [5.3 3.7 1.5 0.2]\n",
      " [5.  3.3 1.4 0.2]\n",
      " [7.  3.2 4.7 1.4]\n",
      " [6.4 3.2 4.5 1.5]\n",
      " [6.9 3.1 4.9 1.5]\n",
      " [5.5 2.3 4.  1.3]\n",
      " [6.5 2.8 4.6 1.5]\n",
      " [5.7 2.8 4.5 1.3]\n",
      " [6.3 3.3 4.7 1.6]\n",
      " [4.9 2.4 3.3 1. ]\n",
      " [6.6 2.9 4.6 1.3]\n",
      " [5.2 2.7 3.9 1.4]\n",
      " [5.  2.  3.5 1. ]\n",
      " [5.9 3.  4.2 1.5]\n",
      " [6.1 2.9 4.7 1.4]\n",
      " [5.6 2.9 3.6 1.3]\n",
      " [6.7 3.1 4.4 1.4]\n",
      " [5.6 3.  4.5 1.5]\n",
      " [5.8 2.7 4.1 1. ]\n",
      " [6.2 2.2 4.5 1.5]\n",
      " [5.9 3.2 4.8 1.8]\n",
      " [6.1 2.8 4.  1.3]\n",
      " [6.3 2.5 4.9 1.5]\n",
      " [6.1 2.8 4.7 1.2]\n",
      " [6.4 2.9 4.3 1.3]\n",
      " [6.6 3.  4.4 1.4]\n",
      " [6.8 2.8 4.8 1.4]\n",
      " [6.  2.9 4.5 1.5]\n",
      " [5.7 2.6 3.5 1. ]\n",
      " [5.5 2.4 3.8 1.1]\n",
      " [5.5 2.4 3.7 1. ]\n",
      " [6.  2.7 5.1 1.6]\n",
      " [5.4 3.  4.5 1.5]\n",
      " [6.7 3.1 4.7 1.5]\n",
      " [6.3 2.3 4.4 1.3]\n",
      " [5.6 3.  4.1 1.3]\n",
      " [5.5 2.5 4.  1.3]\n",
      " [6.1 3.  4.6 1.4]\n",
      " [5.8 2.6 4.  1.2]\n",
      " [5.  2.3 3.3 1. ]\n",
      " [5.6 2.7 4.2 1.3]\n",
      " [5.7 3.  4.2 1.2]\n",
      " [6.2 2.9 4.3 1.3]\n",
      " [5.7 2.8 4.1 1.3]\n",
      " [6.3 3.3 6.  2.5]\n",
      " [7.1 3.  5.9 2.1]\n",
      " [6.3 2.9 5.6 1.8]\n",
      " [6.5 3.  5.8 2.2]\n",
      " [4.9 2.5 4.5 1.7]\n",
      " [6.7 2.5 5.8 1.8]\n",
      " [7.2 3.6 6.1 2.5]\n",
      " [6.4 2.7 5.3 1.9]\n",
      " [6.8 3.  5.5 2.1]\n",
      " [5.7 2.5 5.  2. ]\n",
      " [5.8 2.8 5.1 2.4]\n",
      " [6.4 3.2 5.3 2.3]\n",
      " [6.5 3.  5.5 1.8]\n",
      " [7.7 3.8 6.7 2.2]\n",
      " [7.7 2.6 6.9 2.3]\n",
      " [6.9 3.2 5.7 2.3]\n",
      " [5.6 2.8 4.9 2. ]\n",
      " [7.7 2.8 6.7 2. ]\n",
      " [6.3 2.7 4.9 1.8]\n",
      " [6.7 3.3 5.7 2.1]\n",
      " [7.2 3.2 6.  1.8]\n",
      " [6.2 2.8 4.8 1.8]\n",
      " [6.1 3.  4.9 1.8]\n",
      " [6.4 2.8 5.6 2.1]\n",
      " [7.2 3.  5.8 1.6]\n",
      " [7.4 2.8 6.1 1.9]\n",
      " [6.4 2.8 5.6 2.2]\n",
      " [6.1 2.6 5.6 1.4]\n",
      " [7.7 3.  6.1 2.3]\n",
      " [6.4 3.1 5.5 1.8]\n",
      " [6.  3.  4.8 1.8]\n",
      " [5.8 2.7 5.1 1.9]\n",
      " [6.7 3.  5.2 2.3]\n",
      " [6.5 3.  5.2 2. ]\n",
      " [5.9 3.  5.1 1.8]]\n"
     ]
    }
   ],
   "source": [
    "# create a selection variable which will have 75% True and 25% False values. The size of the selection variable is 150\n",
    "iris_selection = np.random.choice(\n",
    "    [True, False], 150, replace=True, p=[0.75, 0.25])\n",
    "# create an iris dataset using the selection variable\n",
    "data = iris.data[iris_selection]\n",
    "print(data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create a cuml UMAP model \n",
    "fitter = UMAP(n_neighbors=10, min_dist=0.01, verbose=False)\n",
    "# fit the data created the selection variable to the cuml UMAP model created (fitter)\n",
    "fitter.fit(data)\n",
    "# create a new iris dataset by inverting the values of the selection variable (ie. 75% False and 25% True values) \n",
    "new_data = iris.data[~iris_selection]\n",
    "# transform the new data using the previously created embedded space\n",
    "embedding = fitter.transform(new_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9513419913419914\n"
     ]
    }
   ],
   "source": [
    "# calculate the trustworthiness score for the new data created (new_data)\n",
    "trust = trustworthiness(new_data, embedding, 10)\n",
    "print(trust)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}