example of how to freeze embeddings in lightfm by exploiting the accumulated gradient in adagrad

lightfm_freeze-emb_example.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/jtam/tools/lightfm/lightfm/_lightfm_fast.py:9: UserWarning: LightFM was compiled without OpenMP support. Only a single thread will be used.\n",
      "  warnings.warn('LightFM was compiled without OpenMP support. '\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "from lightfm.datasets import fetch_movielens\n",
    "\n",
    "movielens = fetch_movielens()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train = movielens['train']\n",
    "test = movielens['test']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Fitting models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from lightfm import LightFM\n",
    "from lightfm.evaluation import precision_at_k\n",
    "from lightfm.evaluation import auc_score\n",
    "import sys"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_dims = 10\n",
    "model = LightFM(learning_rate=0.05, loss='bpr', random_state=322)\n",
    "\n",
    "# Factor transfer\n",
    "model._initialize(no_components=n_dims,\n",
    "                  no_item_features=train.shape[1],\n",
    "                  no_user_features=train.shape[0],)\n",
    "\n",
    "# Set embeddings for 2 users\n",
    "model.user_embeddings[0,:] = 0.5  # we expect this to be the same\n",
    "model.user_embeddings[1,:] = 0.7  # we expect this to change\n",
    "\n",
    "# Freeze only the first one\n",
    "model.user_embedding_gradients[0,:] = sys.maxsize\n",
    "model.user_bias_gradients[0] = sys.maxsize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<lightfm.lightfm.LightFM at 0x10e58ff60>"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit_partial(train, epochs=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Precision: train 0.59, test 0.10.\n",
      "AUC: train 0.89, test 0.86.\n"
     ]
    }
   ],
   "source": [
    "train_precision = precision_at_k(model, train, k=10).mean()\n",
    "test_precision = precision_at_k(model, test, k=10).mean()\n",
    "\n",
    "train_auc = auc_score(model, train).mean()\n",
    "test_auc = auc_score(model, test).mean()\n",
    "\n",
    "print('Precision: train %.2f, test %.2f.' % (train_precision, test_precision))\n",
    "print('AUC: train %.2f, test %.2f.' % (train_auc, test_auc))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.5       ,  0.5       ,  0.5       ,  0.5       ,  0.5       ,\n",
       "         0.5       ,  0.5       ,  0.5       ,  0.5       ,  0.5       ],\n",
       "       [ 0.09772108, -0.03140765,  1.07019258,  1.09194338,  0.03139753,\n",
       "         0.4900538 ,  1.37998486,  1.55365562,  0.36961207,  0.29496703]], dtype=float32)"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.user_embeddings[:2,:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ -1.78466397e-08,  -1.55268562e+00], dtype=float32)"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.user_biases[:2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [ipykernel_py3]",
   "language": "python",
   "name": "Python [ipykernel_py3]"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.0"
  },
  "widgets": {
   "state": {},
   "version": "1.1.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}