ynet/train_inD_longterm.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "determined-township",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import yaml\n",
    "import argparse\n",
    "import torch\n",
    "from model import YNet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "finished-potential",
   "metadata": {},
   "outputs": [],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e567fbe0",
   "metadata": {},
   "source": [
    "#### Some hyperparameters and settings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "arabic-thickness",
   "metadata": {},
   "outputs": [],
   "source": [
    "CONFIG_FILE_PATH = 'config/inD_longterm.yaml'  # yaml config file containing all the hyperparameters\n",
    "EXPERIMENT_NAME = 'ind_longterm'  # arbitrary name for this experiment\n",
    "DATASET_NAME = 'ind'\n",
    "\n",
    "TRAIN_DATA_PATH = 'data/inD/train.pkl'\n",
    "TRAIN_IMAGE_PATH = 'data/inD/train'\n",
    "VAL_DATA_PATH = 'data/inD/test.pkl'\n",
    "VAL_IMAGE_PATH = 'data/inD/test'\n",
    "OBS_LEN = 5  # in timesteps\n",
    "PRED_LEN = 30  # in timesteps\n",
    "NUM_GOALS = 20  # K_e\n",
    "NUM_TRAJ = 1  # K_a\n",
    "\n",
    "BATCH_SIZE = 8"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2f729e8f",
   "metadata": {},
   "source": [
    "#### Load config file and print hyperparameters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "dangerous-cutting",
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(CONFIG_FILE_PATH) as file:\n",
    "    params = yaml.load(file, Loader=yaml.FullLoader)\n",
    "experiment_name = CONFIG_FILE_PATH.split('.yaml')[0].split('config/')[1]\n",
    "params"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "699a7543",
   "metadata": {},
   "source": [
    "#### Wandb INIT"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "65f560e7",
   "metadata": {},
   "outputs": [],
   "source": [
    "init_wandb(hyperparams.copy(), YNet)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "amber-pressure",
   "metadata": {},
   "source": [
    "#### Load preprocessed Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "german-feature",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train = pd.read_pickle(TRAIN_DATA_PATH)\n",
    "df_val = pd.read_pickle(VAL_DATA_PATH)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "corporate-pharmacy",
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "24dc5d7c",
   "metadata": {},
   "source": [
    "#### Initiate model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "harmful-colleague",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = YNet(obs_len=OBS_LEN, pred_len=PRED_LEN, params=params)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "45e099fe",
   "metadata": {},
   "source": [
    "#### Start training\n",
    "Note, the Val ADE and FDE are without TTST and CWS to save time. Therefore, the numbers will be worse than the final values."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "optional-colleague",
   "metadata": {},
   "outputs": [],
   "source": [
    "model.train(df_train, df_val, params, train_image_path=TRAIN_IMAGE_PATH, val_image_path=VAL_IMAGE_PATH, \n",
    "            experiment_name=EXPERIMENT_NAME, batch_size=BATCH_SIZE, num_goals=NUM_GOALS, num_traj=NUM_TRAJ, \n",
    "            device=None, dataset_name=DATASET_NAME)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "16fe307e",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:PECNet]",
   "language": "python",
   "name": "conda-env-PECNet-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}