Behavioral Input Distribution Training.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Importing the main libraries"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np \n",
    "import matplotlib.pyplot as plt \n",
    "import seaborn as sns\n",
    "sns.set_theme()\n",
    "\n",
    "import torch \n",
    "import torch.nn as nn \n",
    "import torch.optim as optim\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "\n",
    "torch.manual_seed(42)\n",
    "torch.cuda.manual_seed_all(42)\n",
    "torch.set_default_dtype(torch.float32)\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Use GPU if available"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using cuda device\n"
     ]
    }
   ],
   "source": [
    "device = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "print(f\"Using {device} device\")"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Bernstein Polynomials"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils.bernstein_torch import bernstein_coeff_order10_new\n",
    "\n",
    "# Generating P matrix\n",
    "t_fin = 15.0\n",
    "num = 100\n",
    "tot_time = torch.linspace(0, t_fin, num)\n",
    "tot_time_copy = tot_time.reshape(num, 1)\n",
    "P, Pdot, Pddot = bernstein_coeff_order10_new(10, tot_time_copy[0], tot_time_copy[-1], tot_time_copy)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Dataset Loader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import scipy.io as sio\n",
    "\n",
    "# Custom Dataset Loader \n",
    "class TrajDataset(Dataset):\n",
    "    \"\"\"Expert Trajectory Dataset.\"\"\"\n",
    "    def __init__(self, obs, traj):\n",
    "        \n",
    "        # Inputs\n",
    "        self.inp = obs\n",
    "        \n",
    "        # Outputs\n",
    "        self.out = traj\n",
    "        \n",
    "    def __len__(self):\n",
    "        return len(self.inp)\n",
    "    \n",
    "    def __getitem__(self, idx):\n",
    "        \n",
    "        # Inputs\n",
    "        inp = self.inp[idx]\n",
    "        \n",
    "        # Outputs\n",
    "        out = self.out[idx]\n",
    "                 \n",
    "        return torch.tensor(inp).float(), torch.tensor(out).float()\n",
    "\n",
    "# Load the dataset\n",
    "obs = sio.loadmat(\"./dataset/four_lane_dataset.mat\")[\"observations\"]\n",
    "traj = sio.loadmat(\"./dataset/four_lane_dataset.mat\")[\"trajectories\"]\n",
    "\n",
    "# Using PyTorch Dataloader\n",
    "train_dataset = TrajDataset(obs, traj)\n",
    "train_loader = DataLoader(train_dataset, batch_size=8000, shuffle=True, num_workers=0, drop_last=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([8000, 55]) torch.Size([8000, 200])\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Sanity Check\n",
    "\n",
    "for batch_num, (datas) in enumerate(train_loader):\n",
    "    inp, out = datas\n",
    "    \n",
    "    print(inp.shape, out.shape)\n",
    "        \n",
    "    num = 2\n",
    "    \n",
    "    # Sanity Check\n",
    "    x_obs = inp[num].flatten()[5::5]\n",
    "    y_obs = inp[num].flatten()[6::5]\n",
    "     \n",
    "    th = np.linspace(0, 2 * np.pi, 100)\n",
    "    plt.figure(1)\n",
    "    \n",
    "    a_obs, b_obs = 6.0, 3.2\n",
    "    \n",
    "    for i in range(0, 10):\n",
    "        x_ell = x_obs[i] + a_obs * np.cos(th)\n",
    "        y_ell = y_obs[i] + b_obs * np.sin(th)\n",
    "        plt.plot(x_ell, y_ell, '-k', linewidth=1.0)\n",
    "\n",
    "    plt.axis('equal')\n",
    "        \n",
    "    x = out[num].flatten()[0:100]\n",
    "    y = out[num].flatten()[100:]\n",
    "        \n",
    "    plt.plot(x.numpy(), y.numpy(), label=\"Ground Truth\", color=\"red\")\n",
    "    plt.legend()\n",
    "    plt.show()\n",
    "    \n",
    "    break"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### CVAE embedded with a Differentiable Optimization Layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from models.cvae_qp_stack import Encoder, Decoder, Beta_cVAE\n",
    "\n",
    "# Bernstein Polynomial\n",
    "num_batch = train_loader.batch_size\n",
    "P = P.to(device)\n",
    "Pdot = Pdot.to(device)\n",
    "Pddot = Pddot.to(device)\n",
    "P_diag = torch.block_diag(P, P) \n",
    "\n",
    "# Beta-cVAE Inputs\n",
    "enc_inp_dim = 55\n",
    "enc_out_dim = 200\n",
    "dec_inp_dim = enc_inp_dim\n",
    "dec_out_dim = 30 # first 8 behavioral params and 11 lamda x / y's each \n",
    "hidden_dim = 1024 * 2\n",
    "z_dim = 2\n",
    "\n",
    "# Normalization Constant\n",
    "inp_mean, inp_std = -0.1904783993959427, 37.758968353271484\n",
    "\n",
    "# CVAE Encoder-Decoder\n",
    "encoder = Encoder(enc_inp_dim, enc_out_dim, hidden_dim, z_dim)\n",
    "decoder = Decoder(dec_inp_dim, dec_out_dim, hidden_dim, z_dim)\n",
    "model = Beta_cVAE(P, Pdot, Pddot, encoder, decoder, num_batch, inp_mean, inp_std).to(device)"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training Loop"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 1, Train Loss: 94.330, RCL: 82.351, KL: 11.217, Aug: 11.979\n",
      "Epoch: 5, Train Loss: 77.447, RCL: 67.971, KL: 15.853, Aug: 9.421\n",
      "Epoch: 9, Train Loss: 36.301, RCL: 28.268, KL: 17.387, Aug: 7.911\n",
      "Epoch: 13, Train Loss: 33.551, RCL: 25.639, KL: 17.050, Aug: 7.734\n",
      "Epoch: 17, Train Loss: 32.568, RCL: 24.809, KL: 18.169, Aug: 7.505\n",
      "Epoch: 21, Train Loss: 39.545, RCL: 31.212, KL: 18.795, Aug: 8.004\n",
      "Epoch: 25, Train Loss: 30.337, RCL: 23.192, KL: 18.152, Aug: 6.764\n",
      "Epoch: 29, Train Loss: 30.450, RCL: 23.373, KL: 16.888, Aug: 6.663\n",
      "Epoch: 33, Train Loss: 30.716, RCL: 23.441, KL: 17.194, Aug: 6.794\n",
      "Epoch: 37, Train Loss: 28.064, RCL: 20.971, KL: 16.916, Aug: 6.561\n",
      "Epoch: 41, Train Loss: 28.318, RCL: 21.123, KL: 16.661, Aug: 6.612\n",
      "Epoch: 45, Train Loss: 27.870, RCL: 20.633, KL: 16.199, Aug: 6.613\n",
      "Epoch: 49, Train Loss: 27.706, RCL: 20.433, KL: 15.722, Aug: 6.612\n",
      "Epoch: 53, Train Loss: 27.491, RCL: 20.173, KL: 15.232, Aug: 6.625\n",
      "Epoch: 57, Train Loss: 27.790, RCL: 20.509, KL: 14.929, Aug: 6.550\n",
      "Epoch: 61, Train Loss: 27.551, RCL: 20.234, KL: 14.541, Aug: 6.554\n",
      "Epoch: 65, Train Loss: 27.461, RCL: 20.119, KL: 14.512, Aug: 6.530\n",
      "Epoch: 69, Train Loss: 27.174, RCL: 19.790, KL: 14.449, Aug: 6.523\n",
      "Epoch: 73, Train Loss: 27.230, RCL: 19.802, KL: 14.407, Aug: 6.521\n",
      "Epoch: 77, Train Loss: 27.220, RCL: 19.746, KL: 14.359, Aug: 6.520\n",
      "Epoch: 81, Train Loss: 27.366, RCL: 19.850, KL: 14.286, Aug: 6.517\n",
      "Epoch: 85, Train Loss: 27.287, RCL: 19.721, KL: 14.222, Aug: 6.520\n",
      "Epoch: 89, Train Loss: 27.253, RCL: 19.643, KL: 14.142, Aug: 6.521\n",
      "Epoch: 93, Train Loss: 27.285, RCL: 19.635, KL: 14.116, Aug: 6.514\n",
      "Epoch: 97, Train Loss: 27.328, RCL: 19.629, KL: 14.103, Aug: 6.515\n",
      "Epoch: 101, Train Loss: 27.488, RCL: 19.744, KL: 14.092, Aug: 6.512\n",
      "Epoch: 105, Train Loss: 27.509, RCL: 19.720, KL: 14.079, Aug: 6.508\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m/tmp/ipykernel_10781/2693447778.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m     10\u001b[0m     \u001b[0mlosses_train\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mkl_losses\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mrcl_losses\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maug_losses\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     11\u001b[0m     \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtrain\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 12\u001b[0;31m     \u001b[0;32mfor\u001b[0m \u001b[0minp\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mout\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mtrain_loader\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     13\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m         \u001b[0minp\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0minp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mto\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdevice\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/lib/python3.9/site-packages/torch/utils/data/dataloader.py\u001b[0m in \u001b[0;36m__next__\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    528\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_sampler_iter\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    529\u001b[0m                 \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_reset\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 530\u001b[0;31m             \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_next_data\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    531\u001b[0m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_num_yielded\u001b[0m \u001b[0;34m+=\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    532\u001b[0m             \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_dataset_kind\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0m_DatasetKind\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mIterable\u001b[0m \u001b[0;32mand\u001b[0m\u001b[0;31m \u001b[0m\u001b[0;31m\\\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/lib/python3.9/site-packages/torch/utils/data/dataloader.py\u001b[0m in \u001b[0;36m_next_data\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m    568\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_next_data\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    569\u001b[0m         \u001b[0mindex\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_next_index\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m  \u001b[0;31m# may raise StopIteration\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 570\u001b[0;31m         \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_dataset_fetcher\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfetch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mindex\u001b[0m\u001b[0;34m)\u001b[0m  \u001b[0;31m# may raise StopIteration\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    571\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_pin_memory\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    572\u001b[0m             \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_utils\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpin_memory\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mpin_memory\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/lib/python3.9/site-packages/torch/utils/data/_utils/fetch.py\u001b[0m in \u001b[0;36mfetch\u001b[0;34m(self, possibly_batched_index)\u001b[0m\n\u001b[1;32m     47\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mfetch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mpossibly_batched_index\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     48\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mauto_collation\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 49\u001b[0;31m             \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0midx\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0midx\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mpossibly_batched_index\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     50\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     51\u001b[0m             \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mpossibly_batched_index\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/anaconda3/lib/python3.9/site-packages/torch/utils/data/_utils/fetch.py\u001b[0m in \u001b[0;36m<listcomp>\u001b[0;34m(.0)\u001b[0m\n\u001b[1;32m     47\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mfetch\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mpossibly_batched_index\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     48\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mauto_collation\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 49\u001b[0;31m             \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0midx\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0midx\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mpossibly_batched_index\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     50\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     51\u001b[0m             \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdataset\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mpossibly_batched_index\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/tmp/ipykernel_10781/1576544475.py\u001b[0m in \u001b[0;36m__getitem__\u001b[0;34m(self, idx)\u001b[0m\n\u001b[1;32m     23\u001b[0m         \u001b[0mout\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0midx\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     24\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 25\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtensor\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0minp\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfloat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtensor\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfloat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     26\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     27\u001b[0m \u001b[0;31m# Load the dataset\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "epochs = 180\n",
    "step, beta = 0, 3.5\n",
    "optimizer = optim.AdamW(model.parameters(), lr = 2e-4, weight_decay=6e-5)\n",
    "scheduler = optim.lr_scheduler.StepLR(optimizer, step_size = 30, gamma = 0.1)\n",
    "\n",
    "avg_train_loss, avg_rcl_loss, avg_kl_loss, avg_aug_loss = [], [], [], []\n",
    "for epoch in range(epochs):\n",
    "    \n",
    "    # Train Loop\n",
    "    losses_train, kl_losses, rcl_losses, aug_losses = [], [], [], []\n",
    "    model.train()\n",
    "    for inp, out in train_loader:\n",
    "        \n",
    "        inp = inp.to(device)\n",
    "        traj_gt = out.to(device)\n",
    "        \n",
    "        # Ego vehicle states\n",
    "        initial_state_ego = inp[:, 2:6]\n",
    "        initial_state_ego[:, 2:4] = initial_state_ego[:, 0:2]\n",
    "        initial_state_ego[:, 0:2] = 0\n",
    "               \n",
    "        # Prediction & Encoder outputs\n",
    "        traj_sol, res_norm_batch, mu, std= model(inp, traj_gt, initial_state_ego, P_diag)\n",
    "\n",
    "        # Loss Function\n",
    "        aug_loss, KL_loss, RCL_loss, loss = model.cvae_loss(traj_sol, traj_gt, res_norm_batch, mu, std, beta, step)\n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        \n",
    "        losses_train.append(loss.detach().cpu().numpy()) \n",
    "        rcl_losses.append(RCL_loss.detach().cpu().numpy())\n",
    "        kl_losses.append(KL_loss.detach().cpu().numpy())\n",
    "        aug_losses.append(aug_loss.detach().cpu().numpy())\n",
    "\n",
    "    if epoch % 4 == 0:    \n",
    "        print(f\"Epoch: {epoch + 1}, Train Loss: {np.average(losses_train):.3f}, RCL: {np.average(rcl_losses):.3f}, KL: {np.average(kl_losses):.3f}, Aug: {np.average(aug_losses):.3f}\")\n",
    "\n",
    "    step += 0.25 # 0.25\n",
    "    scheduler.step()\n",
    "    avg_train_loss.append(np.average(losses_train)), avg_rcl_loss.append(np.average(rcl_losses)), \\\n",
    "    avg_kl_loss.append(np.average(kl_losses)), avg_aug_loss.append(np.average(aug_losses))"
   ]
  },
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Saving the Weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "torch.save(model.state_dict(), './weights/new_model.pth')"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "base",
   "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": "3.9.7"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "9c562111517df06ec79e2a0b366e2ca57c1da3a77768b6967f7ee5330ff1fbc3"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}