example_torch_gru4rec.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "from experiment_setup import setups"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# download the dataset, the links can be fund in the README\n",
    "dataset_path = \"./datasets/coveo_ecommerce\"\n",
    "model_path = \"./trained_models\""
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Run the preprocess script, specific to the dataset you chose"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- The preprocessing script in general, executes the following steps:\n",
    "    - Loads the raw data, with correct types\n",
    "    - Creates the sessions\n",
    "    - Removes duplicated items. An item is considered as a duplicate if the preceding (based on time) event in the same session contains the exact same item.\n",
    "    - Performes iterative support filtering\n",
    "        - Removes sessions with only one event\n",
    "        - Removes items with less than 5 events\n",
    "        - Until the size of the dataset changes\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%run coveo_preproc.py --path $dataset_path"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Use a specific setup for your dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "params = setups[\"coveo\"][\"params_bprmax\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_path = os.path.join(dataset_path,\"coveo_processed_view_train_full.tsv\")\n",
    "test_path = os.path.join(dataset_path,\"coveo_processed_view_test.tsv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_torch_gru4rec_script(model_name, train_path, test_path, model_path, loss, optim, final_act, layers, batch_size, dropout_p_embed, dropout_p_hidden, learning_rate, sample_alpha, bpreg, n_epochs, n_sample, m):\n",
    "    s_train_full = f\"CUDA_VISIBLE_DEVICES=0 python ./Torch-GRU4Rec/main.py --save_path {model_path}/{model_name} --train_path {train_path} --loss {'nll' if loss =='cross-entropy' else loss} --optimizer {optim} --n_epochs {n_epochs} --embedding_size {layers} --hidden_size {layers} --n_layers {1} --final_act {'softmaxlogit' if final_act=='softmax' else final_act} --batch_size {batch_size} --dropout_p_embed {dropout_p_embed} --dropout_p_hidden {dropout_p_hidden} --lr {learning_rate} --n_sample {n_sample} --sample_alpha {sample_alpha} --bpreg {bpreg}\"\n",
    "    s_test_full = f\"CUDA_VISIBLE_DEVICES=0 python /db_vol/aczapp/sandbox/third_parties/Torch-GRU4Rec/main.py --train_path {train_path} --test_path {test_path} --model_path {model_path}/{model_name}/model_0000{n_epochs-1}.pt --test  --m {m}\"\n",
    "    return s_train_full, s_test_full"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "loss = params[\"loss\"]\n",
    "optim = params[\"optim\"]\n",
    "const_emb = params[\"constrained_embedding\"]\n",
    "embed = params[\"embedding\"]\n",
    "final_act = params[\"final_act\"]\n",
    "layers = params[\"layers\"]\n",
    "batch_size = params[\"batch_size\"]\n",
    "dropout_p_embed = params[\"dropout_p_embed\"]\n",
    "dropout_p_hidden = params[\"dropout_p_hidden\"]\n",
    "learning_rate = params[\"learning_rate\"]\n",
    "momentum = params[\"momentum\"]\n",
    "sample_alpha = params[\"sample_alpha\"]\n",
    "bpreg = params[\"bpreg\"]\n",
    "logq = params[\"logq\"]\n",
    "hidden_act = params[\"hidden_act\"]\n",
    "n_sample = params[\"n_sample\"]\n",
    "n_epochs = 5\n",
    "m = '1 5 10 20'"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train & test the out-of-the-box model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_script_oob, test_script_oob = create_torch_gru4rec_script(model_name='torch_gru4rec_oob_bprmax', train_path=train_path, test_path=test_path, model_path=model_path, loss=loss, optim=optim, final_act=final_act, layers=layers, batch_size=batch_size, dropout_p_embed=dropout_p_embed, dropout_p_hidden=dropout_p_hidden, learning_rate=learning_rate, sample_alpha=sample_alpha, bpreg=bpreg, n_epochs=n_epochs, n_sample=n_sample, m=m)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train the out-of-the-box model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "os.system(train_script_oob)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Test the out-of-the-box model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "os.system(test_script_oob)"
   ]
  },
  {
   "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": "3.8.12"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}