Amazing-Python-Scripts
536 строк · 20.4 Кб
1{2"nbformat": 4,3"nbformat_minor": 0,4"metadata": {5"colab": {6"name": "AutoEncoders.ipynb",7"provenance": [],8"collapsed_sections": [],9"toc_visible": true,10"include_colab_link": true11},12"kernelspec": {13"name": "python3",14"display_name": "Python 3"15},16"accelerator": "GPU"17},18"cells": [19{20"cell_type": "markdown",21"metadata": {22"id": "view-in-github",23"colab_type": "text"24},25"source": [26"<a href=\"https://colab.research.google.com/github/ayush-09/AutoEncoder-Deep-Learning/blob/main/AutoEncoders.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"27]28},29{30"cell_type": "markdown",31"metadata": {32"id": "K4f4JG1gdKqj"33},34"source": [35"#AutoEncoders"36]37},38{39"cell_type": "code",40"metadata": {41"id": "rjOPzue7FCXJ",42"colab": {43"base_uri": "https://localhost:8080/"44},45"outputId": "c488b69b-9cc1-4cd7-b954-0928c41b410d"46},47"source": [48"!wget \"http://files.grouplens.org/datasets/movielens/ml-100k.zip\"\n",49"!unzip ml-100k.zip\n",50"!ls"51],52"execution_count": 1,53"outputs": [54{55"output_type": "stream",56"text": [57"--2021-01-09 14:08:13-- http://files.grouplens.org/datasets/movielens/ml-100k.zip\n",58"Resolving files.grouplens.org (files.grouplens.org)... 128.101.65.152\n",59"Connecting to files.grouplens.org (files.grouplens.org)|128.101.65.152|:80... connected.\n",60"HTTP request sent, awaiting response... 200 OK\n",61"Length: 4924029 (4.7M) [application/zip]\n",62"Saving to: ‘ml-100k.zip’\n",63"\n",64"ml-100k.zip 100%[===================>] 4.70M 18.8MB/s in 0.2s \n",65"\n",66"2021-01-09 14:08:13 (18.8 MB/s) - ‘ml-100k.zip’ saved [4924029/4924029]\n",67"\n",68"Archive: ml-100k.zip\n",69" creating: ml-100k/\n",70" inflating: ml-100k/allbut.pl \n",71" inflating: ml-100k/mku.sh \n",72" inflating: ml-100k/README \n",73" inflating: ml-100k/u.data \n",74" inflating: ml-100k/u.genre \n",75" inflating: ml-100k/u.info \n",76" inflating: ml-100k/u.item \n",77" inflating: ml-100k/u.occupation \n",78" inflating: ml-100k/u.user \n",79" inflating: ml-100k/u1.base \n",80" inflating: ml-100k/u1.test \n",81" inflating: ml-100k/u2.base \n",82" inflating: ml-100k/u2.test \n",83" inflating: ml-100k/u3.base \n",84" inflating: ml-100k/u3.test \n",85" inflating: ml-100k/u4.base \n",86" inflating: ml-100k/u4.test \n",87" inflating: ml-100k/u5.base \n",88" inflating: ml-100k/u5.test \n",89" inflating: ml-100k/ua.base \n",90" inflating: ml-100k/ua.test \n",91" inflating: ml-100k/ub.base \n",92" inflating: ml-100k/ub.test \n",93"ml-100k ml-100k.zip sample_data\n"94],95"name": "stdout"96}97]98},99{100"cell_type": "code",101"metadata": {102"id": "LOly1yfAfTjd",103"colab": {104"base_uri": "https://localhost:8080/"105},106"outputId": "97173fc2-9434-4529-9841-2bb989507c05"107},108"source": [109"!wget \"http://files.grouplens.org/datasets/movielens/ml-1m.zip\"\n",110"!unzip ml-1m.zip\n",111"!ls"112],113"execution_count": 2,114"outputs": [115{116"output_type": "stream",117"text": [118"--2021-01-09 14:08:24-- http://files.grouplens.org/datasets/movielens/ml-1m.zip\n",119"Resolving files.grouplens.org (files.grouplens.org)... 128.101.65.152\n",120"Connecting to files.grouplens.org (files.grouplens.org)|128.101.65.152|:80... connected.\n",121"HTTP request sent, awaiting response... 200 OK\n",122"Length: 5917549 (5.6M) [application/zip]\n",123"Saving to: ‘ml-1m.zip’\n",124"\n",125"ml-1m.zip 100%[===================>] 5.64M 22.1MB/s in 0.3s \n",126"\n",127"2021-01-09 14:08:25 (22.1 MB/s) - ‘ml-1m.zip’ saved [5917549/5917549]\n",128"\n",129"Archive: ml-1m.zip\n",130" creating: ml-1m/\n",131" inflating: ml-1m/movies.dat \n",132" inflating: ml-1m/ratings.dat \n",133" inflating: ml-1m/README \n",134" inflating: ml-1m/users.dat \n",135"ml-100k ml-100k.zip ml-1m ml-1m.zip\tsample_data\n"136],137"name": "stdout"138}139]140},141{142"cell_type": "code",143"metadata": {144"id": "_LvGeU1CeCtg"145},146"source": [147"import numpy as np\n",148"import pandas as pd\n",149"import torch\n",150"import torch.nn as nn\n",151"import torch.nn.parallel\n",152"import torch.optim as optim\n",153"import torch.utils.data\n",154"from torch.autograd import Variable"155],156"execution_count": 3,157"outputs": []158},159{160"cell_type": "code",161"metadata": {162"id": "UJw2p3-Cewo4"163},164"source": [165"# We won't be using this dataset.\n",166"movies = pd.read_csv('ml-1m/movies.dat', sep = '::', header = None, engine = 'python', encoding = 'latin-1')\n",167"users = pd.read_csv('ml-1m/users.dat', sep = '::', header = None, engine = 'python', encoding = 'latin-1')\n",168"ratings = pd.read_csv('ml-1m/ratings.dat', sep = '::', header = None, engine = 'python', encoding = 'latin-1')"169],170"execution_count": 4,171"outputs": []172},173{174"cell_type": "code",175"metadata": {176"id": "2usLKJBEgPE2"177},178"source": [179"training_set = pd.read_csv('ml-100k/u1.base', delimiter = '\\t')\n",180"training_set = np.array(training_set, dtype = 'int')\n",181"test_set = pd.read_csv('ml-100k/u1.test', delimiter = '\\t')\n",182"test_set = np.array(test_set, dtype = 'int')"183],184"execution_count": 6,185"outputs": []186},187{188"cell_type": "code",189"metadata": {190"id": "gPaGZqdniC5m"191},192"source": [193"nb_users = int(max(max(training_set[:, 0], ), max(test_set[:, 0])))\n",194"nb_movies = int(max(max(training_set[:, 1], ), max(test_set[:, 1])))"195],196"execution_count": 7,197"outputs": []198},199{200"cell_type": "code",201"metadata": {202"id": "-wASs2YFiDaa"203},204"source": [205"def convert(data):\n",206" new_data = []\n",207" for id_users in range(1, nb_users + 1):\n",208" id_movies = data[:, 1] [data[:, 0] == id_users]\n",209" id_ratings = data[:, 2] [data[:, 0] == id_users]\n",210" ratings = np.zeros(nb_movies)\n",211" ratings[id_movies - 1] = id_ratings\n",212" new_data.append(list(ratings))\n",213" return new_data\n",214"training_set = convert(training_set)\n",215"test_set = convert(test_set)"216],217"execution_count": 8,218"outputs": []219},220{221"cell_type": "code",222"metadata": {223"id": "TwD-KD8yiEEw"224},225"source": [226"training_set = torch.FloatTensor(training_set)\n",227"test_set = torch.FloatTensor(test_set)"228],229"execution_count": 9,230"outputs": []231},232{233"cell_type": "code",234"metadata": {235"id": "oU2nyh76iE6M"236},237"source": [238"class SAE(nn.Module):\n",239" def __init__(self, ):\n",240" super(SAE, self).__init__()\n",241" self.fc1 = nn.Linear(nb_movies, 20)\n",242" self.fc2 = nn.Linear(20, 10)\n",243" self.fc3 = nn.Linear(10, 20)\n",244" self.fc4 = nn.Linear(20, nb_movies)\n",245" self.activation = nn.Sigmoid()\n",246" def forward(self, x):\n",247" x = self.activation(self.fc1(x))\n",248" x = self.activation(self.fc2(x))\n",249" x = self.activation(self.fc3(x))\n",250" x = self.fc4(x)\n",251" return x\n",252"sae = SAE()\n",253"criterion = nn.MSELoss()\n",254"optimizer = optim.RMSprop(sae.parameters(), lr = 0.01, weight_decay = 0.5)"255],256"execution_count": 10,257"outputs": []258},259{260"cell_type": "code",261"metadata": {262"id": "FEz9hRaciFTs",263"colab": {264"base_uri": "https://localhost:8080/"265},266"outputId": "8ffae3d0-f74c-47af-c046-fbe8a5fec0b6"267},268"source": [269"nb_epoch = 200\n",270"for epoch in range(1, nb_epoch + 1):\n",271" train_loss = 0\n",272" s = 0.\n",273" for id_user in range(nb_users):\n",274" input = Variable(training_set[id_user]).unsqueeze(0)\n",275" target = input.clone()\n",276" if torch.sum(target.data > 0) > 0:\n",277" output = sae(input)\n",278" target.require_grad = False\n",279" output[target == 0] = 0\n",280" loss = criterion(output, target)\n",281" mean_corrector = nb_movies/float(torch.sum(target.data > 0) + 1e-10)\n",282" loss.backward()\n",283" train_loss += np.sqrt(loss.data*mean_corrector)\n",284" s += 1.\n",285" optimizer.step()\n",286" print('epoch: '+str(epoch)+'loss: '+ str(train_loss/s))"287],288"execution_count": 11,289"outputs": [290{291"output_type": "stream",292"text": [293"epoch: 1loss: tensor(1.7710)\n",294"epoch: 2loss: tensor(1.0968)\n",295"epoch: 3loss: tensor(1.0533)\n",296"epoch: 4loss: tensor(1.0386)\n",297"epoch: 5loss: tensor(1.0307)\n",298"epoch: 6loss: tensor(1.0267)\n",299"epoch: 7loss: tensor(1.0237)\n",300"epoch: 8loss: tensor(1.0220)\n",301"epoch: 9loss: tensor(1.0206)\n",302"epoch: 10loss: tensor(1.0196)\n",303"epoch: 11loss: tensor(1.0188)\n",304"epoch: 12loss: tensor(1.0185)\n",305"epoch: 13loss: tensor(1.0178)\n",306"epoch: 14loss: tensor(1.0176)\n",307"epoch: 15loss: tensor(1.0174)\n",308"epoch: 16loss: tensor(1.0170)\n",309"epoch: 17loss: tensor(1.0166)\n",310"epoch: 18loss: tensor(1.0166)\n",311"epoch: 19loss: tensor(1.0163)\n",312"epoch: 20loss: tensor(1.0160)\n",313"epoch: 21loss: tensor(1.0161)\n",314"epoch: 22loss: tensor(1.0160)\n",315"epoch: 23loss: tensor(1.0157)\n",316"epoch: 24loss: tensor(1.0156)\n",317"epoch: 25loss: tensor(1.0157)\n",318"epoch: 26loss: tensor(1.0155)\n",319"epoch: 27loss: tensor(1.0155)\n",320"epoch: 28loss: tensor(1.0150)\n",321"epoch: 29loss: tensor(1.0134)\n",322"epoch: 30loss: tensor(1.0120)\n",323"epoch: 31loss: tensor(1.0096)\n",324"epoch: 32loss: tensor(1.0085)\n",325"epoch: 33loss: tensor(1.0048)\n",326"epoch: 34loss: tensor(1.0045)\n",327"epoch: 35loss: tensor(1.0007)\n",328"epoch: 36loss: tensor(0.9990)\n",329"epoch: 37loss: tensor(0.9970)\n",330"epoch: 38loss: tensor(0.9964)\n",331"epoch: 39loss: tensor(0.9932)\n",332"epoch: 40loss: tensor(0.9912)\n",333"epoch: 41loss: tensor(0.9876)\n",334"epoch: 42loss: tensor(0.9896)\n",335"epoch: 43loss: tensor(0.9856)\n",336"epoch: 44loss: tensor(0.9904)\n",337"epoch: 45loss: tensor(0.9861)\n",338"epoch: 46loss: tensor(0.9854)\n",339"epoch: 47loss: tensor(0.9880)\n",340"epoch: 48loss: tensor(0.9873)\n",341"epoch: 49loss: tensor(0.9877)\n",342"epoch: 50loss: tensor(0.9880)\n",343"epoch: 51loss: tensor(0.9817)\n",344"epoch: 52loss: tensor(0.9830)\n",345"epoch: 53loss: tensor(0.9797)\n",346"epoch: 54loss: tensor(0.9768)\n",347"epoch: 55loss: tensor(0.9728)\n",348"epoch: 56loss: tensor(0.9808)\n",349"epoch: 57loss: tensor(0.9754)\n",350"epoch: 58loss: tensor(0.9756)\n",351"epoch: 59loss: tensor(0.9712)\n",352"epoch: 60loss: tensor(0.9708)\n",353"epoch: 61loss: tensor(0.9720)\n",354"epoch: 62loss: tensor(0.9683)\n",355"epoch: 63loss: tensor(0.9652)\n",356"epoch: 64loss: tensor(0.9625)\n",357"epoch: 65loss: tensor(0.9638)\n",358"epoch: 66loss: tensor(0.9625)\n",359"epoch: 67loss: tensor(0.9599)\n",360"epoch: 68loss: tensor(0.9605)\n",361"epoch: 69loss: tensor(0.9611)\n",362"epoch: 70loss: tensor(0.9588)\n",363"epoch: 71loss: tensor(0.9564)\n",364"epoch: 72loss: tensor(0.9566)\n",365"epoch: 73loss: tensor(0.9554)\n",366"epoch: 74loss: tensor(0.9574)\n",367"epoch: 75loss: tensor(0.9542)\n",368"epoch: 76loss: tensor(0.9544)\n",369"epoch: 77loss: tensor(0.9526)\n",370"epoch: 78loss: tensor(0.9506)\n",371"epoch: 79loss: tensor(0.9498)\n",372"epoch: 80loss: tensor(0.9484)\n",373"epoch: 81loss: tensor(0.9472)\n",374"epoch: 82loss: tensor(0.9473)\n",375"epoch: 83loss: tensor(0.9461)\n",376"epoch: 84loss: tensor(0.9461)\n",377"epoch: 85loss: tensor(0.9450)\n",378"epoch: 86loss: tensor(0.9442)\n",379"epoch: 87loss: tensor(0.9432)\n",380"epoch: 88loss: tensor(0.9428)\n",381"epoch: 89loss: tensor(0.9423)\n",382"epoch: 90loss: tensor(0.9425)\n",383"epoch: 91loss: tensor(0.9407)\n",384"epoch: 92loss: tensor(0.9414)\n",385"epoch: 93loss: tensor(0.9408)\n",386"epoch: 94loss: tensor(0.9402)\n",387"epoch: 95loss: tensor(0.9396)\n",388"epoch: 96loss: tensor(0.9393)\n",389"epoch: 97loss: tensor(0.9386)\n",390"epoch: 98loss: tensor(0.9384)\n",391"epoch: 99loss: tensor(0.9381)\n",392"epoch: 100loss: tensor(0.9390)\n",393"epoch: 101loss: tensor(0.9381)\n",394"epoch: 102loss: tensor(0.9378)\n",395"epoch: 103loss: tensor(0.9370)\n",396"epoch: 104loss: tensor(0.9372)\n",397"epoch: 105loss: tensor(0.9360)\n",398"epoch: 106loss: tensor(0.9363)\n",399"epoch: 107loss: tensor(0.9354)\n",400"epoch: 108loss: tensor(0.9353)\n",401"epoch: 109loss: tensor(0.9346)\n",402"epoch: 110loss: tensor(0.9355)\n",403"epoch: 111loss: tensor(0.9347)\n",404"epoch: 112loss: tensor(0.9349)\n",405"epoch: 113loss: tensor(0.9337)\n",406"epoch: 114loss: tensor(0.9335)\n",407"epoch: 115loss: tensor(0.9332)\n",408"epoch: 116loss: tensor(0.9334)\n",409"epoch: 117loss: tensor(0.9331)\n",410"epoch: 118loss: tensor(0.9333)\n",411"epoch: 119loss: tensor(0.9327)\n",412"epoch: 120loss: tensor(0.9328)\n",413"epoch: 121loss: tensor(0.9322)\n",414"epoch: 122loss: tensor(0.9319)\n",415"epoch: 123loss: tensor(0.9317)\n",416"epoch: 124loss: tensor(0.9317)\n",417"epoch: 125loss: tensor(0.9318)\n",418"epoch: 126loss: tensor(0.9311)\n",419"epoch: 127loss: tensor(0.9313)\n",420"epoch: 128loss: tensor(0.9310)\n",421"epoch: 129loss: tensor(0.9313)\n",422"epoch: 130loss: tensor(0.9312)\n",423"epoch: 131loss: tensor(0.9307)\n",424"epoch: 132loss: tensor(0.9305)\n",425"epoch: 133loss: tensor(0.9298)\n",426"epoch: 134loss: tensor(0.9300)\n",427"epoch: 135loss: tensor(0.9299)\n",428"epoch: 136loss: tensor(0.9299)\n",429"epoch: 137loss: tensor(0.9293)\n",430"epoch: 138loss: tensor(0.9294)\n",431"epoch: 139loss: tensor(0.9287)\n",432"epoch: 140loss: tensor(0.9285)\n",433"epoch: 141loss: tensor(0.9281)\n",434"epoch: 142loss: tensor(0.9284)\n",435"epoch: 143loss: tensor(0.9279)\n",436"epoch: 144loss: tensor(0.9278)\n",437"epoch: 145loss: tensor(0.9274)\n",438"epoch: 146loss: tensor(0.9278)\n",439"epoch: 147loss: tensor(0.9275)\n",440"epoch: 148loss: tensor(0.9273)\n",441"epoch: 149loss: tensor(0.9270)\n",442"epoch: 150loss: tensor(0.9269)\n",443"epoch: 151loss: tensor(0.9260)\n",444"epoch: 152loss: tensor(0.9264)\n",445"epoch: 153loss: tensor(0.9259)\n",446"epoch: 154loss: tensor(0.9257)\n",447"epoch: 155loss: tensor(0.9252)\n",448"epoch: 156loss: tensor(0.9256)\n",449"epoch: 157loss: tensor(0.9249)\n",450"epoch: 158loss: tensor(0.9250)\n",451"epoch: 159loss: tensor(0.9245)\n",452"epoch: 160loss: tensor(0.9249)\n",453"epoch: 161loss: tensor(0.9238)\n",454"epoch: 162loss: tensor(0.9249)\n",455"epoch: 163loss: tensor(0.9244)\n",456"epoch: 164loss: tensor(0.9242)\n",457"epoch: 165loss: tensor(0.9234)\n",458"epoch: 166loss: tensor(0.9253)\n",459"epoch: 167loss: tensor(0.9234)\n",460"epoch: 168loss: tensor(0.9237)\n",461"epoch: 169loss: tensor(0.9227)\n",462"epoch: 170loss: tensor(0.9235)\n",463"epoch: 171loss: tensor(0.9227)\n",464"epoch: 172loss: tensor(0.9225)\n",465"epoch: 173loss: tensor(0.9217)\n",466"epoch: 174loss: tensor(0.9224)\n",467"epoch: 175loss: tensor(0.9225)\n",468"epoch: 176loss: tensor(0.9247)\n",469"epoch: 177loss: tensor(0.9219)\n",470"epoch: 178loss: tensor(0.9210)\n",471"epoch: 179loss: tensor(0.9210)\n",472"epoch: 180loss: tensor(0.9211)\n",473"epoch: 181loss: tensor(0.9209)\n",474"epoch: 182loss: tensor(0.9210)\n",475"epoch: 183loss: tensor(0.9204)\n",476"epoch: 184loss: tensor(0.9208)\n",477"epoch: 185loss: tensor(0.9205)\n",478"epoch: 186loss: tensor(0.9206)\n",479"epoch: 187loss: tensor(0.9202)\n",480"epoch: 188loss: tensor(0.9201)\n",481"epoch: 189loss: tensor(0.9195)\n",482"epoch: 190loss: tensor(0.9200)\n",483"epoch: 191loss: tensor(0.9200)\n",484"epoch: 192loss: tensor(0.9199)\n",485"epoch: 193loss: tensor(0.9195)\n",486"epoch: 194loss: tensor(0.9193)\n",487"epoch: 195loss: tensor(0.9198)\n",488"epoch: 196loss: tensor(0.9195)\n",489"epoch: 197loss: tensor(0.9181)\n",490"epoch: 198loss: tensor(0.9188)\n",491"epoch: 199loss: tensor(0.9183)\n",492"epoch: 200loss: tensor(0.9221)\n"493],494"name": "stdout"495}496]497},498{499"cell_type": "code",500"metadata": {501"id": "5ztvzYRtiGCz",502"colab": {503"base_uri": "https://localhost:8080/",504"height": 34505},506"outputId": "d0e8ea8b-9ac4-40e5-a19a-7fcfc6934d61"507},508"source": [509"test_loss = 0\n",510"s = 0.\n",511"for id_user in range(nb_users):\n",512" input = Variable(training_set[id_user]).unsqueeze(0)\n",513" target = Variable(test_set[id_user]).unsqueeze(0)\n",514" if torch.sum(target.data > 0) > 0:\n",515" output = sae(input)\n",516" target.require_grad = False\n",517" output[target == 0] = 0\n",518" loss = criterion(output, target)\n",519" mean_corrector = nb_movies/float(torch.sum(target.data > 0) + 1e-10)\n",520" test_loss += np.sqrt(loss.data*mean_corrector)\n",521" s += 1.\n",522"print('test loss: '+str(test_loss/s))"523],524"execution_count": null,525"outputs": [526{527"output_type": "stream",528"text": [529"test loss: tensor(0.9681)\n"530],531"name": "stdout"532}533]534}535]536}