PriyaB272 · Apr 27, 2018 · Apr 27, 2018 · Apr 27, 2018 · Apr 27, 2018 · Apr 27, 2018
diff --git a/Code/BookLook_RBM.ipynb b/Code/BookLook_RBM.ipynb
@@ -0,0 +1 @@
+{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"name":"RBM_bookLook.ipynb","version":"0.3.2","views":{},"default_view":{},"provenance":[]},"kernelspec":{"display_name":"Python 2","language":"python","name":"python2"}},"cells":[{"metadata":{"id":"bZ7J1PgBxzG0","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["## Implementing RBM\n","import numpy as np\n","import pandas as pd\n","import torch\n","import torch.nn as nn     # for building neural network\n","import torch.nn.parallel     # for parallel computing\n","import torch.optim as optim    # for optimisation\n","import torch.utils.data   \n","from torch.autograd import Variable  # for stocastic gradient descent"],"execution_count":0,"outputs":[]},{"metadata":{"id":"jrkPw02JxzG5","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["df = pd.read_csv('BX-Book-Ratings.csv', names=[\"User-ID\", \"ISBN\", \"Book-Rating\"], delimiter=\";\")\n","\n","df_mat = df.as_matrix()[1:]\n","\n","df_mat[:,0] = df_mat[:,0].astype('int64')\n","df_mat[:,2] = df_mat[:,2].astype('int64')\n","\n","usercount = (len(np.unique(df_mat[:,0])))\n","bookcount = (len(np.unique(df_mat[:,1])))\n","\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"rlLnseH2xzG9","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["rating_matrix = np.ndarray(shape=(10000,10000), dtype = 'int64')\n","rating_matrix = np.zeros(shape=(10000,10000))\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"cwfyYLS5xzHB","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["rating_matrix_test = np.zeros(shape=(10000,10000))\n","rating_matrix_test = np.ndarray(shape=(10000,10000), dtype = 'int64')"],"execution_count":0,"outputs":[]},{"metadata":{"id":"PCNIhKBZxzHE","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["bookid = 0\n","usernum = 0\n","bookmap = {}\n","usermap = {}\n","useridslist = []\n","for k in range(0,500000):\n","    userid = df_mat[k,0]\n","    isbn = df_mat[k,1]\n","    if isbn not in bookmap:\n","        if bookid < 10000:\n","            bookmap[isbn] = bookid\n","            if userid not in usermap:\n","                if usernum < 10000:\n","                    usermap[userid]=usernum\n","                    rating_matrix[usernum, bookid] = df_mat[k,2]\n","                    usernum+=1\n","            else:\n","                rating_matrix[usermap[userid], bookid] = df_mat[k,2]\n","            bookid = bookid+1\n","    else:\n","        if userid not in usermap:\n","            if usernum < 10000:\n","                usermap[userid]=usernum\n","                rating_matrix[usernum, bookmap[isbn]] = df_mat[k,2]\n","                usernum+=1\n","        else:\n","            rating_matrix[usermap[userid], bookmap[isbn]] = df_mat[k,2]\n","            "],"execution_count":0,"outputs":[]},{"metadata":{"id":"4xH0LdcrxzHI","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["usertestmap = {}\n","booktestmap = {}\n","uid = 0 \n","iid = 0"],"execution_count":0,"outputs":[]},{"metadata":{"id":"aZ6XecQexzHL","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["bookid = 0\n","usernum = 0\n","booktestmap = {}\n","usertestmap = {}\n","for k in range(700000, 1000000):\n","    userid = df_mat[k,0]\n","    isbn = df_mat[k,1]\n","    if isbn not in booktestmap:\n","        if bookid < 10000:\n","            booktestmap[isbn] = bookid\n","            if userid not in usertestmap:\n","                if usernum < 10000:\n","                    usertestmap[userid]=usernum\n","                    rating_matrix_test[usernum, bookid] = df_mat[k,2]\n","                    usernum+=1\n","            else:\n","                rating_matrix_test[usertestmap[userid], bookid] = df_mat[k,2]\n","            bookid = bookid+1\n","    else:\n","        if userid not in usertestmap:\n","            if usernum < 10000:\n","                usertestmap[userid]=usernum\n","                rating_matrix_test[usernum, booktestmap[isbn]] = df_mat[k,2]\n","                usernum+=1\n","        else:\n","            rating_matrix_test[usertestmap[userid], booktestmap[isbn]] = df_mat[k,2]"],"execution_count":0,"outputs":[]},{"metadata":{"id":"_Am_JqZVxzHP","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}},"outputId":"40ad82ed-e0d4-413f-efe8-c7649891cc18"},"cell_type":"code","source":["print rating_matrix_test"],"execution_count":0,"outputs":[{"output_type":"stream","text":["[[10  0  0 ...,  0  0  0]\n"," [ 0  0  0 ...,  0  0  0]\n"," [ 0  0  0 ...,  0  0  0]\n"," ..., \n"," [ 0  0  0 ...,  0  0  0]\n"," [ 0  0  0 ...,  0  0  0]\n"," [ 0  0  0 ...,  0  0  0]]\n"],"name":"stdout"}]},{"metadata":{"id":"96dOUDj-xzHW","colab_type":"text"},"cell_type":"markdown","source":[""]},{"metadata":{"id":"z5gS7bTNxzHX","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# converting the data into torch tensors\n","rating_matrix = torch.FloatTensor(rating_matrix)\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"V4G4T_-pxzHa","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# converting ratings to binary: rating > 6 is liked = 1, otw not liked\n","# the original 0's in the training set are given a value of -1\n","\n","rating_matrix[rating_matrix == 0] = -1\n","\n","rating_matrix[rating_matrix == 1] = 0\n","rating_matrix[rating_matrix == 2] = 0\n","rating_matrix[rating_matrix == 3] = 0\n","rating_matrix[rating_matrix == 4] = 0\n","rating_matrix[rating_matrix == 5] = 0\n","rating_matrix[rating_matrix == 6] = 0\n","rating_matrix[rating_matrix == 7] = 0\n","\n","# converting all ratings > 7 to 1\n","rating_matrix[rating_matrix >= 8] = 1\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"c1HgINrWxzHe","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["rating_matrix_test = torch.FloatTensor(rating_matrix_test)\n","\n","rating_matrix_test[rating_matrix_test == 0] = -1\n","rating_matrix_test[rating_matrix_test == 1] = 0\n","rating_matrix_test[rating_matrix_test == 2] = 0\n","rating_matrix_test[rating_matrix_test == 3] = 0\n","rating_matrix_test[rating_matrix_test == 4] = 0\n","rating_matrix_test[rating_matrix_test == 5] = 0\n","rating_matrix_test[rating_matrix_test == 6] = 0\n","rating_matrix_test[rating_matrix_test == 7] = 0\n","rating_matrix_test[rating_matrix_test >= 8] = 1"],"execution_count":0,"outputs":[]},{"metadata":{"id":"UnO2D3bcxzHi","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# creating the RBM neural network\n","class RBM():\n","    def __init__(self, nv, nh):\n","        self.W = torch.randn(nh, nv)   #initialising weights in torch\n","        self.a = torch.randn(1, nh)   # to initialise the RBM object\n","        self.b = torch.randn(1, nv)   # for bias\n","    \n","    # use gibbs sampling to improve log likelihood gradient\n","    # sample_h : sample prob of hidden nodes, given visible nodes\n","    def sample_h(self, x):\n","        wx = torch.mm(x, self.W.t())  # multiply w(tensor weights) and x(visible neurons)\n","        activ = wx + self.a.expand_as(wx) # to add bias to every row of neuron\n","        prob_h_given_v = torch.sigmoid(activ)\n","        return prob_h_given_v, torch.bernoulli(prob_h_given_v) # returns prob and neurons activated by this sampling\n","    \n","    # sample v: prob of visible nodes, given hidden nodes\n","    # estimating prob of visible nodes\n","    def sample_v(self,y):\n","        wy = torch.mm(y, self.W)  # multiply w(tensor weights) and x(visible neurons)\n","        activ = wy + self.b.expand_as(wy) # to add bias to every row of neuron\n","        prob_v_given_h = torch.sigmoid(activ)\n","        return prob_v_given_h, torch.bernoulli(prob_v_given_h) # returns prob and neurons activated by this sampling\n","    \n","    def train(self,v0, vk, ph0, phk):  # Using k-step contrastive divergence\n","    #v0:input vec having ratings of all movies by 1 user, vk: visible nodes obtained after k iterations, \n","    #ph0: prob of hidden nodes= 1 in first iteration , phk: prob of hidden nodes after k sampling\n","        self.W += torch.mm(v0.t(),ph0) - torch.mm(vk.t(), phk)\n","        self.b += torch.sum((v0-vk),0) \n","        self.a += torch.sum((ph0-phk),0)\n","        \n","    "],"execution_count":0,"outputs":[]},{"metadata":{"id":"RCIu1c8mxzHn","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}},"outputId":"e424c1ed-a7bd-4d1b-c038-79a0c13f113a"},"cell_type":"code","source":["nv = len(rating_matrix[0]) # no. of visible nodes\n","print nv\n","nh = 100 # number of features we want to detect\n","\n","# batch size: to update weights after several iterations and not afer each one\n","batch_size = 1000\n","rbm = RBM(nv,nh)\n"],"execution_count":0,"outputs":[{"output_type":"stream","text":["10000\n"],"name":"stdout"}]},{"metadata":{"id":"01rmpelpxzHt","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}},"outputId":"36173cc5-6945-4c67-cfd0-1b8ca0f372ee"},"cell_type":"code","source":["n_iter = 5\n","for iter in range(0, n_iter):\n","    train_loss = 0\n","    rms_train_loss = 0\n","    counter = 0.\n","    for user in range(0, 10000 - batch_size, batch_size):\n","        vk = rating_matrix[user:user+batch_size]# op of gibbs sampling afte k iterations of random walk\n","        v0 = rating_matrix[user:user+batch_size]\n","        ph0,_ = rbm.sample_h(v0)\n","        for k in range(10):  # for k steps of contrastive div\n","            _,hk = rbm.sample_h(vk)\n","            _,vk = rbm.sample_v(hk)\n","            vk[v0<0]= v0[v0<0] # to eliminate -1 ratings\n","        phk,_ = rbm.sample_h(vk)\n","        rbm.train(v0,vk,ph0,phk)\n","        train_loss += torch.mean(torch.abs(v0[v0>=0] - vk[v0>=0])) #MAE\n","        rms_train_loss += torch.mean((v0[v0>=0] - vk[v0>=0])**2) #RMSE\n","        #train_loss += torch.mean(torch.abs(v0[v0>=0] - vk[v0>=0])) #MAE\n","        counter += 1.\n","    print \"MAE  on training for iter #\"+str(iter)+\" is: \"+str(train_loss/counter)\n","    print \"RMSE on training for iter #\"+str(iter)+\" is: \"+str((rms_train_loss/counter)**0.5)\n","    print \"\"\n","        \n","        \n","            \n","        "],"execution_count":0,"outputs":[{"output_type":"stream","text":["MAE  on training for iter #0 is: 0.380275072065\n","RMSE on training for iter #0 is: 0.616664472842\n","\n","MAE  on training for iter #1 is: 0.377912293285\n","RMSE on training for iter #1 is: 0.614745714328\n","\n","MAE  on training for iter #2 is: 0.375525298138\n","RMSE on training for iter #2 is: 0.612801189733\n","\n","MAE  on training for iter #3 is: 0.370339020544\n","RMSE on training for iter #3 is: 0.608554862394\n","\n","MAE  on training for iter #4 is: 0.371679191732\n","RMSE on training for iter #4 is: 0.609654977616\n","\n"],"name":"stdout"}]},{"metadata":{"id":"k9-SXoYLxzHy","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}},"outputId":"8c7c8dd2-4a33-4ebe-cafe-bad0c3f33d4c"},"cell_type":"code","source":["\n","test_loss = 0\n","counter = 0.\n","mae_test_loss = 0\n","rmse_test_loss = 0\n","\n","for user in range(10000):\n","    v = rating_matrix[user:user+1]\n","    vt = rating_matrix_test[user:user+1]\n","    if len(vt[vt>=0]) > 0:\n","        _,h = rbm.sample_h(v)\n","        _,v = rbm.sample_v(h)\n","    if len(vt[vt>=0]) > 0 and len(v[vt>=0]) > 0:\n","        mae_test_loss += torch.mean(torch.abs(vt[vt>=0] - v[vt>=0])) #MAE\n","        rmse_test_loss += torch.mean((vt[vt>=0] - v[vt>=0])**2) #RMSE\n","    counter += 1.\n","print \"MAE  on testing is: \"+str(mae_test_loss/counter)\n","print \"RMSE on testing is: \"+str((rmse_test_loss/counter)**0.5)\n","        \n"],"execution_count":0,"outputs":[{"output_type":"stream","text":["MAE  on training for iter #4 is: 0.347591827294\n","RMSE on training for iter #4 is: 0.589569187877\n"],"name":"stdout"}]},{"metadata":{"id":"yt5V-q8ExzH2","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":[""],"execution_count":0,"outputs":[]}]}
diff --git a/Code/BookLook_SVD.ipynb b/Code/BookLook_SVD.ipynb
diff --git a/Code/BookLook_final.ipynb b/Code/BookLook_final.ipynb
diff --git a/README.md b/README.md
@@ -1,74 +1 @@
-# Particle Jekyll Theme
 
-![](./particle.jpg)
-
-This is a simple and minimalist template for Jekyll designed for developers that want to show of their portfolio.
-
-The Theme features:
-
-- Gulp
-- SASS
-- Sweet Scroll
-- Particle.js
-- BrowserSync
-- Font Awesome and Devicon icons
-- Google Analytics
-- Info Customization
-
-## Basic Setup
-
-1. [Install Jekyll](http://jekyllrb.com)
-2. Fork the [Particle Theme](https://github.com/nrandecker/particle/fork)
-3. Clone the repo you just forked.
-4. Edit `_config.yml` to personalize your site.
-
-## Site and User Settings
-
-You have to fill some informations on `_config.yml` to customize your site.
-
-```
-# Site settings
-description: A blog about lorem ipsum dolor sit amet
-baseurl: "" # the subpath of your site, e.g. /blog/
-url: "http://localhost:3000" # the base hostname & protocol for your site
-
-# User settings
-username: Lorem Ipsum
-user_description: Anon Developer at Lorem Ipsum Dolor
-user_title: Anon Developer
-email: anon@anon.com
-twitter_username: lorem_ipsum
-github_username:  lorem_ipsum
-gplus_username:  lorem_ipsum
-```
-
-**Don't forget to change your url before you deploy your site!**
-
-## Color and Particle Customization
-- Color Customization
-  - Edit the sass variables
-- Particle Customization
-  - Edit the json data in particle function in app.js
-  - Refer to [Particle.js](https://github.com/VincentGarreau/particles.js/) for help
-
-## Running the blog in local
-
-In order to compile the assets and run Jekyll on local you need to follow those steps:
-
-- Install [NodeJS](https://nodejs.org/)
-- Run `npm install`
-- Run `gulp`
-
-## Questions
-
-Having any issues file a [GitHub Issue](https://github.com/nrandecker/particle/issues/new).
-
-## License
-
-This theme is free and open source software, distributed under the The MIT License. So feel free to use this Jekyll theme anyway you want.
-
-## Credits
-
-This theme was partially designed with the inspiration from these fine folks
-- [Willian Justen](https://github.com/willianjusten/will-jekyll-template)
-- [Vincent Garreau](https://github.com/VincentGarreau/particles.js/)
diff --git a/_includes/codelogistics.html b/_includes/codelogistics.html
@@ -0,0 +1,12 @@
+<div class="user-details">
+ <h1>Code Logistics</h1>
+</div>
+<div class="user">
+  <div class="tech">
+    <p align="center">The ipynb files with the code can be found in the below location. There are 3 separate files in the location.
+    The file BookLook_final.ipynb is the file that has the LHSCM implememntation. The BookLook_SVD.ipynb is the file that has 
+    the SVD model implementation and BookLook_RBM has the Neural network model implementation.</br>
+    <a href="https://github.com/fnumegha/BookLook/tree/master/Code" target="_blank">https://github.com/fnumegha/BookLook/tree/master/Code</a>
+    </p>
+   </div>
+  </div>
diff --git a/_includes/footer.html b/_includes/footer.html
@@ -1,6 +1,6 @@
 <footer class="footer">
     <p>&copy; {{site.username}}</p>
-    <p>Build with Jekyll and <span class="love">❤</span> by <a href="https://github.com/nrandecker">Nathan Randecker</a></p>
+    <p>A CSCE 670 adventure. Built with <span class="love">❤</span></p>
 </footer>
 <script src="//cdn.jsdelivr.net/particles.js/2.0.0/particles.min.js"></script>
 <script src="{{ "/assets/js/sweet-scroll.min.js" | prepend: site.baseurl }}"></script>

diff --git a/_includes/header.html b/_includes/header.html
@@ -6,10 +6,12 @@ <h1>
       </h1>
       <div class="header-links">
         <a class="link" href="#about" data-scroll>About</a>
-        <a class="link" href="#relatedwork" data-scroll>Related Work</a>
+        <a class="link" href="#projects" data-scroll>Methodology</a>
         <a class="link" href="#lhscm" data-scroll>LHSCM Method</a>
         <a class="link" href="#evaluation" data-scroll>Evaluation</a>
         <a class="link" href="#rbm" data-scroll>RBM</a>
+        <a class="link" href="#rbm" data-scroll>Code</a>
+        <a class="link" href="#references" data-scroll>References</a>
       </div>
   </div>
   <a class="down" href="#about" data-scroll><i class="icon fa fa-chevron-down" aria-hidden="true"></i></a>

diff --git a/_includes/projects.html b/_includes/projects.html
@@ -3,11 +3,12 @@ <h1> Methodology </h1>
 </div>
 <div class="user">
     <div class="tech">
-      <p>To take advantage of the dislikes of the user, we plan to use the method of 'Love-Hate Square Counting Method' proposed in
+      <p align = "center">To take advantage of the dislikes of the user, we plan to use the method of 'Love-Hate Square Counting Method' proposed in
         [1] Let's take an example to explain this! Suppose Cav is friends with Trump and Parisa. Now, Cav hates 3 books which Trump 
         also hates & Cav loves 3 books which Parisa also loves. Now if we know that Trump likes "Twilight" and Parisa likes "Life of 
         Pie", then , is it more likely for Cav to like "Twilight" or "Life of Pie"? 
         This is exactly the question we are trying to answer with BookLook! </p>
    </div>
 </div>
-
+<div class="images">
+        <img alt="LoveHate" src="{{ "/assets/img/LoveHate.jpg" | prepend: site.baseurl }}" /></div> 
diff --git a/_includes/results.html b/_includes/results.html
@@ -1,5 +1,5 @@
 <div class="user-details">
-  <h1>Results</h1>
+  <h1>Evaluation</h1>
 
 </div>
 

diff --git a/_layouts/default.html b/_layouts/default.html
@@ -27,6 +27,9 @@
       <section id="rbm">
         {% include NeuralNets.html %}
       </section>
+      <section id="code">
+         {% include codelogistics.html %}
+    </section>
       <section id="references">
         {% include references.html %}
       </section>

diff --git a/assets/img/LoveHate.jpg b/assets/img/LoveHate.jpg
Original file line number	Diff line number	Diff line change
		@@ -0,0 +1 @@
		{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"name":"RBM_bookLook.ipynb","version":"0.3.2","views":{},"default_view":{},"provenance":[]},"kernelspec":{"display_name":"Python 2","language":"python","name":"python2"}},"cells":[{"metadata":{"id":"bZ7J1PgBxzG0","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["## Implementing RBM\n","import numpy as np\n","import pandas as pd\n","import torch\n","import torch.nn as nn # for building neural network\n","import torch.nn.parallel # for parallel computing\n","import torch.optim as optim # for optimisation\n","import torch.utils.data \n","from torch.autograd import Variable # for stocastic gradient descent"],"execution_count":0,"outputs":[]},{"metadata":{"id":"jrkPw02JxzG5","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["df = pd.read_csv('BX-Book-Ratings.csv', names=[\"User-ID\", \"ISBN\", \"Book-Rating\"], delimiter=\";\")\n","\n","df_mat = df.as_matrix()[1:]\n","\n","df_mat[:,0] = df_mat[:,0].astype('int64')\n","df_mat[:,2] = df_mat[:,2].astype('int64')\n","\n","usercount = (len(np.unique(df_mat[:,0])))\n","bookcount = (len(np.unique(df_mat[:,1])))\n","\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"rlLnseH2xzG9","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["rating_matrix = np.ndarray(shape=(10000,10000), dtype = 'int64')\n","rating_matrix = np.zeros(shape=(10000,10000))\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"cwfyYLS5xzHB","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["rating_matrix_test = np.zeros(shape=(10000,10000))\n","rating_matrix_test = np.ndarray(shape=(10000,10000), dtype = 'int64')"],"execution_count":0,"outputs":[]},{"metadata":{"id":"PCNIhKBZxzHE","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["bookid = 0\n","usernum = 0\n","bookmap = {}\n","usermap = {}\n","useridslist = []\n","for k in range(0,500000):\n"," userid = df_mat[k,0]\n"," isbn = df_mat[k,1]\n"," if isbn not in bookmap:\n"," if bookid < 10000:\n"," bookmap[isbn] = bookid\n"," if userid not in usermap:\n"," if usernum < 10000:\n"," usermap[userid]=usernum\n"," rating_matrix[usernum, bookid] = df_mat[k,2]\n"," usernum+=1\n"," else:\n"," rating_matrix[usermap[userid], bookid] = df_mat[k,2]\n"," bookid = bookid+1\n"," else:\n"," if userid not in usermap:\n"," if usernum < 10000:\n"," usermap[userid]=usernum\n"," rating_matrix[usernum, bookmap[isbn]] = df_mat[k,2]\n"," usernum+=1\n"," else:\n"," rating_matrix[usermap[userid], bookmap[isbn]] = df_mat[k,2]\n"," "],"execution_count":0,"outputs":[]},{"metadata":{"id":"4xH0LdcrxzHI","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["usertestmap = {}\n","booktestmap = {}\n","uid = 0 \n","iid = 0"],"execution_count":0,"outputs":[]},{"metadata":{"id":"aZ6XecQexzHL","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["bookid = 0\n","usernum = 0\n","booktestmap = {}\n","usertestmap = {}\n","for k in range(700000, 1000000):\n"," userid = df_mat[k,0]\n"," isbn = df_mat[k,1]\n"," if isbn not in booktestmap:\n"," if bookid < 10000:\n"," booktestmap[isbn] = bookid\n"," if userid not in usertestmap:\n"," if usernum < 10000:\n"," usertestmap[userid]=usernum\n"," rating_matrix_test[usernum, bookid] = df_mat[k,2]\n"," usernum+=1\n"," else:\n"," rating_matrix_test[usertestmap[userid], bookid] = df_mat[k,2]\n"," bookid = bookid+1\n"," else:\n"," if userid not in usertestmap:\n"," if usernum < 10000:\n"," usertestmap[userid]=usernum\n"," rating_matrix_test[usernum, booktestmap[isbn]] = df_mat[k,2]\n"," usernum+=1\n"," else:\n"," rating_matrix_test[usertestmap[userid], booktestmap[isbn]] = df_mat[k,2]"],"execution_count":0,"outputs":[]},{"metadata":{"id":"_Am_JqZVxzHP","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}},"outputId":"40ad82ed-e0d4-413f-efe8-c7649891cc18"},"cell_type":"code","source":["print rating_matrix_test"],"execution_count":0,"outputs":[{"output_type":"stream","text":["[[10 0 0 ..., 0 0 0]\n"," [ 0 0 0 ..., 0 0 0]\n"," [ 0 0 0 ..., 0 0 0]\n"," ..., \n"," [ 0 0 0 ..., 0 0 0]\n"," [ 0 0 0 ..., 0 0 0]\n"," [ 0 0 0 ..., 0 0 0]]\n"],"name":"stdout"}]},{"metadata":{"id":"96dOUDj-xzHW","colab_type":"text"},"cell_type":"markdown","source":[""]},{"metadata":{"id":"z5gS7bTNxzHX","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# converting the data into torch tensors\n","rating_matrix = torch.FloatTensor(rating_matrix)\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"V4G4T_-pxzHa","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# converting ratings to binary: rating > 6 is liked = 1, otw not liked\n","# the original 0's in the training set are given a value of -1\n","\n","rating_matrix[rating_matrix == 0] = -1\n","\n","rating_matrix[rating_matrix == 1] = 0\n","rating_matrix[rating_matrix == 2] = 0\n","rating_matrix[rating_matrix == 3] = 0\n","rating_matrix[rating_matrix == 4] = 0\n","rating_matrix[rating_matrix == 5] = 0\n","rating_matrix[rating_matrix == 6] = 0\n","rating_matrix[rating_matrix == 7] = 0\n","\n","# converting all ratings > 7 to 1\n","rating_matrix[rating_matrix >= 8] = 1\n"],"execution_count":0,"outputs":[]},{"metadata":{"id":"c1HgINrWxzHe","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["rating_matrix_test = torch.FloatTensor(rating_matrix_test)\n","\n","rating_matrix_test[rating_matrix_test == 0] = -1\n","rating_matrix_test[rating_matrix_test == 1] = 0\n","rating_matrix_test[rating_matrix_test == 2] = 0\n","rating_matrix_test[rating_matrix_test == 3] = 0\n","rating_matrix_test[rating_matrix_test == 4] = 0\n","rating_matrix_test[rating_matrix_test == 5] = 0\n","rating_matrix_test[rating_matrix_test == 6] = 0\n","rating_matrix_test[rating_matrix_test == 7] = 0\n","rating_matrix_test[rating_matrix_test >= 8] = 1"],"execution_count":0,"outputs":[]},{"metadata":{"id":"UnO2D3bcxzHi","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":["# creating the RBM neural network\n","class RBM():\n"," def __init__(self, nv, nh):\n"," self.W = torch.randn(nh, nv) #initialising weights in torch\n"," self.a = torch.randn(1, nh) # to initialise the RBM object\n"," self.b = torch.randn(1, nv) # for bias\n"," \n"," # use gibbs sampling to improve log likelihood gradient\n"," # sample_h : sample prob of hidden nodes, given visible nodes\n"," def sample_h(self, x):\n"," wx = torch.mm(x, self.W.t()) # multiply w(tensor weights) and x(visible neurons)\n"," activ = wx + self.a.expand_as(wx) # to add bias to every row of neuron\n"," prob_h_given_v = torch.sigmoid(activ)\n"," return prob_h_given_v, torch.bernoulli(prob_h_given_v) # returns prob and neurons activated by this sampling\n"," \n"," # sample v: prob of visible nodes, given hidden nodes\n"," # estimating prob of visible nodes\n"," def sample_v(self,y):\n"," wy = torch.mm(y, self.W) # multiply w(tensor weights) and x(visible neurons)\n"," activ = wy + self.b.expand_as(wy) # to add bias to every row of neuron\n"," prob_v_given_h = torch.sigmoid(activ)\n"," return prob_v_given_h, torch.bernoulli(prob_v_given_h) # returns prob and neurons activated by this sampling\n"," \n"," def train(self,v0, vk, ph0, phk): # Using k-step contrastive divergence\n"," #v0:input vec having ratings of all movies by 1 user, vk: visible nodes obtained after k iterations, \n"," #ph0: prob of hidden nodes= 1 in first iteration , phk: prob of hidden nodes after k sampling\n"," self.W += torch.mm(v0.t(),ph0) - torch.mm(vk.t(), phk)\n"," self.b += torch.sum((v0-vk),0) \n"," self.a += torch.sum((ph0-phk),0)\n"," \n"," "],"execution_count":0,"outputs":[]},{"metadata":{"id":"RCIu1c8mxzHn","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}},"outputId":"e424c1ed-a7bd-4d1b-c038-79a0c13f113a"},"cell_type":"code","source":["nv = len(rating_matrix[0]) # no. of visible nodes\n","print nv\n","nh = 100 # number of features we want to detect\n","\n","# batch size: to update weights after several iterations and not afer each one\n","batch_size = 1000\n","rbm = RBM(nv,nh)\n"],"execution_count":0,"outputs":[{"output_type":"stream","text":["10000\n"],"name":"stdout"}]},{"metadata":{"id":"01rmpelpxzHt","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}},"outputId":"36173cc5-6945-4c67-cfd0-1b8ca0f372ee"},"cell_type":"code","source":["n_iter = 5\n","for iter in range(0, n_iter):\n"," train_loss = 0\n"," rms_train_loss = 0\n"," counter = 0.\n"," for user in range(0, 10000 - batch_size, batch_size):\n"," vk = rating_matrix[user:user+batch_size]# op of gibbs sampling afte k iterations of random walk\n"," v0 = rating_matrix[user:user+batch_size]\n"," ph0,_ = rbm.sample_h(v0)\n"," for k in range(10): # for k steps of contrastive div\n"," _,hk = rbm.sample_h(vk)\n"," _,vk = rbm.sample_v(hk)\n"," vk[v0<0]= v0[v0<0] # to eliminate -1 ratings\n"," phk,_ = rbm.sample_h(vk)\n"," rbm.train(v0,vk,ph0,phk)\n"," train_loss += torch.mean(torch.abs(v0[v0>=0] - vk[v0>=0])) #MAE\n"," rms_train_loss += torch.mean((v0[v0>=0] - vk[v0>=0])2) #RMSE\n"," #train_loss += torch.mean(torch.abs(v0[v0>=0] - vk[v0>=0])) #MAE\n"," counter += 1.\n"," print \"MAE on training for iter #\"+str(iter)+\" is: \"+str(train_loss/counter)\n"," print \"RMSE on training for iter #\"+str(iter)+\" is: \"+str((rms_train_loss/counter)0.5)\n"," print \"\"\n"," \n"," \n"," \n"," "],"execution_count":0,"outputs":[{"output_type":"stream","text":["MAE on training for iter #0 is: 0.380275072065\n","RMSE on training for iter #0 is: 0.616664472842\n","\n","MAE on training for iter #1 is: 0.377912293285\n","RMSE on training for iter #1 is: 0.614745714328\n","\n","MAE on training for iter #2 is: 0.375525298138\n","RMSE on training for iter #2 is: 0.612801189733\n","\n","MAE on training for iter #3 is: 0.370339020544\n","RMSE on training for iter #3 is: 0.608554862394\n","\n","MAE on training for iter #4 is: 0.371679191732\n","RMSE on training for iter #4 is: 0.609654977616\n","\n"],"name":"stdout"}]},{"metadata":{"id":"k9-SXoYLxzHy","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}},"outputId":"8c7c8dd2-4a33-4ebe-cafe-bad0c3f33d4c"},"cell_type":"code","source":["\n","test_loss = 0\n","counter = 0.\n","mae_test_loss = 0\n","rmse_test_loss = 0\n","\n","for user in range(10000):\n"," v = rating_matrix[user:user+1]\n"," vt = rating_matrix_test[user:user+1]\n"," if len(vt[vt>=0]) > 0:\n"," _,h = rbm.sample_h(v)\n"," _,v = rbm.sample_v(h)\n"," if len(vt[vt>=0]) > 0 and len(v[vt>=0]) > 0:\n"," mae_test_loss += torch.mean(torch.abs(vt[vt>=0] - v[vt>=0])) #MAE\n"," rmse_test_loss += torch.mean((vt[vt>=0] - v[vt>=0])2) #RMSE\n"," counter += 1.\n","print \"MAE on testing is: \"+str(mae_test_loss/counter)\n","print \"RMSE on testing is: \"+str((rmse_test_loss/counter)0.5)\n"," \n"],"execution_count":0,"outputs":[{"output_type":"stream","text":["MAE on training for iter #4 is: 0.347591827294\n","RMSE on training for iter #4 is: 0.589569187877\n"],"name":"stdout"}]},{"metadata":{"id":"yt5V-q8ExzH2","colab_type":"code","colab":{"autoexec":{"startup":false,"wait_interval":0}}},"cell_type":"code","source":[""],"execution_count":0,"outputs":[]}]}