example.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "toc": true
   },
   "source": [
    "<h1>Table of Contents<span class=\"tocSkip\"></span></h1>\n",
    "<div class=\"toc\"><ul class=\"toc-item\"><li><span><a href=\"#Load-the-Data-&amp;-Train-on-a-small-subset\" data-toc-modified-id=\"Load-the-Data-&amp;-Train-on-a-small-subset-1\"><span class=\"toc-item-num\">1&nbsp;&nbsp;</span>Load the Data &amp; Train on a small subset</a></span></li><li><span><a href=\"#Full-Training\" data-toc-modified-id=\"Full-Training-2\"><span class=\"toc-item-num\">2&nbsp;&nbsp;</span>Full Training</a></span><ul class=\"toc-item\"><li><span><a href=\"#Evaluate-using-the-trained-models\" data-toc-modified-id=\"Evaluate-using-the-trained-models-2.1\"><span class=\"toc-item-num\">2.1&nbsp;&nbsp;</span>Evaluate using the trained models</a></span></li><li><span><a href=\"#1.Load-the-Common-Parameters\" data-toc-modified-id=\"1.Load-the-Common-Parameters-2.2\"><span class=\"toc-item-num\">2.2&nbsp;&nbsp;</span>1.Load the Common Parameters</a></span></li><li><span><a href=\"#2.-Evaluation(TOP1-Loss)\" data-toc-modified-id=\"2.-Evaluation(TOP1-Loss)-2.3\"><span class=\"toc-item-num\">2.3&nbsp;&nbsp;</span>2. Evaluation(TOP1 Loss)</a></span></li><li><span><a href=\"#3.-Evaluation(BPR-Loss)\" data-toc-modified-id=\"3.-Evaluation(BPR-Loss)-2.4\"><span class=\"toc-item-num\">2.4&nbsp;&nbsp;</span>3. Evaluation(BPR Loss)</a></span></li><li><span><a href=\"#4.-Evaluation(CrossEntropyLoss)\" data-toc-modified-id=\"4.-Evaluation(CrossEntropyLoss)-2.5\"><span class=\"toc-item-num\">2.5&nbsp;&nbsp;</span>4. Evaluation(CrossEntropyLoss)</a></span></li></ul></li></ul></div>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load the Data & Train on a small subset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-02-27T22:25:36.070715Z",
     "start_time": "2018-02-27T22:25:25.615253Z"
    }
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from pathlib import Path\n",
    "from modules.data import SessionDataset\n",
    "\n",
    "PATH_HOME = Path.home()\n",
    "PATH_PROJ = PATH_HOME/'pyGRU4REC' \n",
    "PATH_DATA = PATH_PROJ/'data'\n",
    "PATH_MODEL = PATH_PROJ/'models'\n",
    "train = 'train.tsv'\n",
    "test = 'test.tsv'\n",
    "PATH_TRAIN = PATH_DATA / train\n",
    "PATH_TEST = PATH_DATA / test\n",
    "n_samples = -1\n",
    "n_samples = 100000\n",
    "train_dataset = SessionDataset(PATH_TRAIN, n_samples=n_samples)\n",
    "test_dataset = SessionDataset(PATH_TEST, n_samples=n_samples, itemmap=train_dataset.itemmap)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Train on a small subset of data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from modules.model import GRU4REC\n",
    "import torch\n",
    "\n",
    "input_size = len(train_dataset.items)\n",
    "hidden_size = 100\n",
    "num_layers = 1\n",
    "output_size = input_size\n",
    "batch_size = 50\n",
    "\n",
    "optimizer_type = 'Adagrad'\n",
    "lr = .01\n",
    "weight_decay = 0\n",
    "momentum = 0\n",
    "eps = 1e-6\n",
    "\n",
    "loss_type = 'TOP1'\n",
    "\n",
    "n_epochs = 10\n",
    "use_cuda = True\n",
    "\n",
    "torch.manual_seed(7)\n",
    "\n",
    "model = GRU4REC(input_size, hidden_size, output_size,\n",
    "                num_layers=num_layers,\n",
    "                batch_size=batch_size,\n",
    "                optimizer_type=optimizer_type,\n",
    "                lr=lr,\n",
    "                weight_decay=weight_decay,\n",
    "                momentum=momentum,\n",
    "                eps=eps,\n",
    "                loss_type=loss_type,\n",
    "                use_cuda=use_cuda)\n",
    "\n",
    "model_name = 'GRU4REC'\n",
    "model.train(train_dataset, n_epochs=n_epochs, model_name=model_name, save=False)\n",
    "model.test(test_dataset)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Full Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-02-27T05:29:50.087721Z",
     "start_time": "2018-02-27T05:29:37.345345Z"
    }
   },
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import torch\n",
    "from modules.data import SessionDataset\n",
    "from modules.model import GRU4REC\n",
    "\n",
    "\n",
    "PATH_HOME = Path.home()\n",
    "PATH_PROJ = PATH_HOME/'pyGRU4REC' \n",
    "PATH_DATA = PATH_PROJ/'data'\n",
    "PATH_MODEL = PATH_PROJ/'models'\n",
    "train = 'train.tsv'\n",
    "test = 'test.tsv'\n",
    "PATH_TRAIN = PATH_DATA / train\n",
    "PATH_TEST = PATH_DATA / test\n",
    "\n",
    "train_dataset = SessionDataset(PATH_TRAIN)\n",
    "test_dataset = SessionDataset(PATH_TEST, itemmap=train_dataset.itemmap)\n",
    "\n",
    "input_size = len(train_dataset.items)\n",
    "hidden_size = 100\n",
    "num_layers = 1\n",
    "output_size = input_size\n",
    "batch_size = 50\n",
    "\n",
    "optimizer_type = 'Adagrad'\n",
    "lr = .01\n",
    "weight_decay = 0\n",
    "momentum = 0\n",
    "eps = 1e-6\n",
    "\n",
    "loss_type = 'TOP1'\n",
    "\n",
    "n_epochs = 10\n",
    "use_cuda = True\n",
    "\n",
    "torch.manual_seed(7)\n",
    "\n",
    "model = GRU4REC(input_size, hidden_size, output_size,\n",
    "                num_layers=num_layers,\n",
    "                batch_size=batch_size,\n",
    "                optimizer_type=optimizer_type,\n",
    "                lr=lr,\n",
    "                weight_decay=weight_decay,\n",
    "                momentum=momentum,\n",
    "                eps=eps,\n",
    "                loss_type=loss_type,\n",
    "                use_cuda=use_cuda)\n",
    "\n",
    "model_name = 'GRU4REC'\n",
    "model.train(train_dataset, n_epochs=n_epochs, model_name=model_name, save=False)\n",
    "model.test(test_dataset)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluate using the trained models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-02-27T05:44:13.202111Z",
     "start_time": "2018-02-27T05:44:05.731675Z"
    }
   },
   "outputs": [],
   "source": [
    "from pathlib import Path\n",
    "from modules.data import SessionDataset\n",
    "\n",
    "PATH_HOME = Path.home()\n",
    "PATH_PROJ = PATH_HOME/'pyGRU4REC' \n",
    "PATH_DATA = PATH_PROJ/'data'\n",
    "PATH_MODEL = PATH_PROJ/'models'\n",
    "train = 'train.tsv'\n",
    "test = 'test.tsv'\n",
    "PATH_TRAIN = PATH_DATA / train\n",
    "PATH_TEST = PATH_DATA / test\n",
    "train_dataset = SessionDataset(PATH_TRAIN)\n",
    "test_dataset = SessionDataset(PATH_TEST, itemmap=train_dataset.itemmap)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1.Load the Common Parameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-02-27T05:44:13.441073Z",
     "start_time": "2018-02-27T05:44:13.316621Z"
    }
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "from modules.layer import GRU\n",
    "from modules.model import GRU4REC\n",
    "\n",
    "input_size = len(train_dataset.items)\n",
    "output_size = input_size\n",
    "hidden_size = 100\n",
    "num_layers = 1\n",
    "\n",
    "use_cuda = True\n",
    "time_sort = False\n",
    "\n",
    "optimizer_type = 'Adagrad'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Evaluation(TOP1 Loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'GRU4REC_TOP1_Adagrad_0.01_epoch5'\n",
    "model_file = PATH_MODEL/model_name\n",
    "\n",
    "loss_type = 'TOP1'\n",
    "lr = 0.01\n",
    "\n",
    "dropout_hidden = .5\n",
    "dropout_input = 0\n",
    "batch_size = 50\n",
    "momentum = 0\n",
    "\n",
    "gru = GRU(input_size, hidden_size, output_size,\n",
    "          num_layers = num_layers,\n",
    "          dropout_input = dropout_input,\n",
    "          dropout_hidden = dropout_hidden,\n",
    "          batch_size = batch_size,\n",
    "          use_cuda = use_cuda)\n",
    "\n",
    "gru.load_state_dict(torch.load(model_file))\n",
    "\n",
    "model = GRU4REC(input_size, hidden_size, output_size,\n",
    "                num_layers = num_layers,\n",
    "                dropout_input = dropout_input,\n",
    "                dropout_hidden = dropout_hidden,\n",
    "                batch_size = batch_size,\n",
    "                use_cuda = use_cuda,\n",
    "                loss_type = loss_type,\n",
    "                optimizer_type = optimizer_type,\n",
    "                lr=lr,\n",
    "                momentum=momentum,\n",
    "                time_sort=time_sort,\n",
    "                pretrained=gru)\n",
    "\n",
    "k = 20\n",
    "model.test(test_dataset, k=k)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Evaluation(BPR Loss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'GRU4REC_BPR_Adagrad_0.05_epoch5'\n",
    "model_file = PATH_MODEL/model_name\n",
    "\n",
    "loss_type = 'BPR'\n",
    "lr = 0.05\n",
    "\n",
    "dropout_hidden = .2\n",
    "dropout_input = 0\n",
    "batch_size = 50\n",
    "momentum = 0.2\n",
    "\n",
    "gru = GRU(input_size, hidden_size, output_size,\n",
    "          num_layers = num_layers,\n",
    "          dropout_input = dropout_input,\n",
    "          dropout_hidden = dropout_hidden,\n",
    "          batch_size = batch_size,\n",
    "          use_cuda = use_cuda)\n",
    "\n",
    "gru.load_state_dict(torch.load(model_file))\n",
    "\n",
    "model = GRU4REC(input_size, hidden_size, output_size,\n",
    "                num_layers = num_layers,\n",
    "                dropout_input = dropout_input,\n",
    "                dropout_hidden = dropout_hidden,\n",
    "                batch_size = batch_size,\n",
    "                use_cuda = use_cuda,\n",
    "                loss_type = loss_type,\n",
    "                optimizer_type = optimizer_type,\n",
    "                lr=lr,\n",
    "                momentum=momentum,\n",
    "                time_sort=time_sort,\n",
    "                pretrained=gru)\n",
    "\n",
    "k = 20\n",
    "model.test(test_dataset, k=k)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Evaluation(CrossEntropyLoss)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'GRU4REC_CrossEntropy_Adagrad_0.01_epoch5'\n",
    "model_file = PATH_MODEL/model_name\n",
    "\n",
    "loss_type = 'CrossEntropy'\n",
    "lr = 0.01\n",
    "\n",
    "dropout_hidden = 0\n",
    "dropout_input = 0\n",
    "batch_size = 500\n",
    "momentum = 0\n",
    "\n",
    "gru = GRU(input_size, hidden_size, output_size,\n",
    "          num_layers = num_layers,\n",
    "          dropout_input = dropout_input,\n",
    "          dropout_hidden = dropout_hidden,\n",
    "          batch_size = batch_size,\n",
    "          use_cuda = use_cuda)\n",
    "\n",
    "gru.load_state_dict(torch.load(model_file))\n",
    "\n",
    "model = GRU4REC(input_size, hidden_size, output_size,\n",
    "                num_layers = num_layers,\n",
    "                dropout_input = dropout_input,\n",
    "                dropout_hidden = dropout_hidden,\n",
    "                batch_size = batch_size,\n",
    "                use_cuda = use_cuda,\n",
    "                loss_type = loss_type,\n",
    "                optimizer_type = optimizer_type,\n",
    "                lr=lr,\n",
    "                momentum=momentum,\n",
    "                time_sort=time_sort,\n",
    "                pretrained=gru)\n",
    "\n",
    "k = 20\n",
    "model.test(test_dataset, k=k)"
   ]
  }
 ],
 "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": "3.6.8"
  },
  "latex_envs": {
   "LaTeX_envs_menu_present": true,
   "autoclose": false,
   "autocomplete": true,
   "bibliofile": "biblio.bib",
   "cite_by": "apalike",
   "current_citInitial": 1,
   "eqLabelWithNumbers": true,
   "eqNumInitial": 1,
   "hotkeys": {
    "equation": "Ctrl-E",
    "itemize": "Ctrl-I"
   },
   "labels_anchors": false,
   "latex_user_defs": false,
   "report_style_numbering": false,
   "user_envs_cfg": false
  },
  "toc": {
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": true,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": true
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "position": {
    "height": "761px",
    "left": "1497px",
    "right": "20px",
    "top": "121px",
    "width": "354px"
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}