test_cnmp.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Device : cuda:0\n"
     ]
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "from models.cnmp import CNMP\n",
    "from data.data_generators import *\n",
    "\n",
    "import torch\n",
    "\n",
    "torch.set_float32_matmul_precision('high')\n",
    "\n",
    "def get_free_gpu():\n",
    "    gpu_util = []\n",
    "    for i in range(torch.cuda.device_count()):\n",
    "        torch.cuda.set_device(i)  # Switch GPU\n",
    "#        gpu_util.append((i, torch.cuda.memory_stats()['reserved_bytes.all.current'] / (1024 ** 2)))\n",
    "        gpu_util.append((i, torch.cuda.utilization()))\n",
    "    gpu_util.sort(key=lambda x: x[1])\n",
    "    return gpu_util[0][0]\n",
    "\n",
    "if torch.cuda.is_available():\n",
    "    available_gpu = get_free_gpu()\n",
    "    if available_gpu == 0:\n",
    "        device = torch.device(\"cuda:0\")\n",
    "    else:\n",
    "        device = torch.device(f\"cuda:{available_gpu}\")\n",
    "else:\n",
    "    device = torch.device(\"cpu\")\n",
    "\n",
    "print(\"Device :\", device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([32, 200, 1]) torch.Size([32, 200, 1]) cuda:0\n",
      "torch.Size([8, 200, 1]) torch.Size([8, 200, 1])\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Data generation\n",
    "\n",
    "dx, dy = 1, 1\n",
    "t_steps = 200\n",
    "num_train_per_class, num_val_per_class = 32, 8\n",
    "num_demos, num_val = 32, 8\n",
    "num_classes = 1\n",
    "n_max, m_max = 10, 10\n",
    "\n",
    "x = torch.linspace(0, 1, t_steps).view(-1, 1)\n",
    "\n",
    "generator_functions = [generate_sin]\n",
    "y = []\n",
    "for i in range(num_train_per_class + num_val_per_class):\n",
    "    noise = (torch.randn((t_steps, 1))-0.5)*0.015\n",
    "    y.append(generator_functions[0](x)+noise)\n",
    "\n",
    "colors = [sns.color_palette('tab10')[0], sns.color_palette('tab10')[1], sns.color_palette('tab10')[2], sns.color_palette('tab10')[3]]\n",
    "sns.set_palette('tab10')\n",
    "\n",
    "for i, y_i in enumerate(y):\n",
    "    if i < num_train_per_class:\n",
    "        plt.plot(y_i, alpha=0.2, c=colors[0])\n",
    "    else:\n",
    "        plt.plot(y_i, alpha=0.2, c=colors[1])\n",
    "\n",
    "x = x.unsqueeze(0).repeat(len(y), 1, 1).to(device)\n",
    "y = torch.stack(y, dim=0).to(device)\n",
    "\n",
    "vx, vy = x[num_train_per_class*num_classes:], y[num_train_per_class*num_classes:]\n",
    "x, y = x[:num_train_per_class*num_classes], y[:num_train_per_class*num_classes]\n",
    "\n",
    "print(x.shape, y.shape, x.device)\n",
    "print(vx.shape, vy.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "8706\n"
     ]
    }
   ],
   "source": [
    "batch_size = 8\n",
    "\n",
    "model = CNMP(encoder_hidden_dims=[64,64], decoder_hidden_dims=[64,64], batch_size=batch_size, device=device)\n",
    "optimizer = torch.optim.Adam(lr=1e-4, params=model.parameters())\n",
    "\n",
    "pytorch_total_params = sum(p.numel() for p in model.parameters())\n",
    "print(pytorch_total_params)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "obs = torch.zeros((batch_size, n_max, dx+dy), dtype=torch.float32, device=device)\n",
    "tar_x = torch.zeros((batch_size, m_max, dx), dtype=torch.float32, device=device)\n",
    "tar_y = torch.zeros((batch_size, m_max, dy), dtype=torch.float32, device=device)\n",
    "obs_mask = torch.zeros((batch_size, n_max), dtype=torch.bool, device=device)\n",
    "tar_mask = torch.zeros((batch_size, m_max), dtype=torch.bool, device=device)\n",
    "\n",
    "def prepare_masked_batch(t: list, traj_ids: list):\n",
    "    obs.fill_(0)\n",
    "    tar_x.fill_(0)\n",
    "    tar_y.fill_(0)\n",
    "    obs_mask.fill_(False)\n",
    "    tar_mask.fill_(False)\n",
    "\n",
    "    for i, traj_id in enumerate(traj_ids):\n",
    "        traj = t[traj_id]\n",
    "\n",
    "        n = torch.randint(1, n_max, (1,)).item()\n",
    "        m = torch.randint(1, m_max, (1,)).item()\n",
    "\n",
    "        permuted_ids = torch.randperm(t_steps)\n",
    "        n_ids = permuted_ids[:n]\n",
    "        m_ids = permuted_ids[n:n+m]\n",
    "        \n",
    "        obs[i, :n, :dx] = traj[n_ids]\n",
    "        obs[i, :n, dx:] = (n_ids/t_steps).unsqueeze(1)\n",
    "        obs_mask[i, :n] = True\n",
    "        \n",
    "        tar_x[i, :m] = traj[m_ids]\n",
    "        tar_y[i, :m] = (m_ids/t_steps).unsqueeze(1)\n",
    "        tar_mask[i, :m] = True\n",
    "\n",
    "val_obs = torch.zeros((batch_size, n_max, dx+dy), dtype=torch.float32, device=device)\n",
    "val_tar_x = torch.zeros((batch_size, t_steps, dx), dtype=torch.float32, device=device)\n",
    "val_tar_y = torch.zeros((batch_size, t_steps, dy), dtype=torch.float32, device=device)\n",
    "val_obs_mask = torch.zeros((batch_size, n_max), dtype=torch.bool, device=device)\n",
    "\n",
    "def prepare_masked_val_batch(t: list, traj_ids: list):\n",
    "    val_obs.fill_(0)\n",
    "    val_tar_x.fill_(0)\n",
    "    val_tar_y.fill_(0)\n",
    "    val_obs_mask.fill_(False)\n",
    "\n",
    "    for i, traj_id in enumerate(traj_ids):\n",
    "        traj = t[traj_id]\n",
    "\n",
    "        n = torch.randint(1, n_max, (1,)).item()\n",
    "\n",
    "        permuted_ids = torch.randperm(t_steps)\n",
    "        n_ids = permuted_ids[:n]\n",
    "        m_ids = torch.arange(t_steps)\n",
    "        \n",
    "        val_obs[i, :n, :dx] = traj[n_ids]\n",
    "        val_obs[i, :n, dx:] = (n_ids/t_steps).unsqueeze(1)\n",
    "        val_obs_mask[i, :n] = True\n",
    "        \n",
    "        val_tar_x[i] = traj[m_ids]\n",
    "        val_tar_y[i] = (m_ids/t_steps).unsqueeze(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "New best: 0.13594388961791992\n",
      "Epoch: 0, CNMP Losses: 0.0007603584975004196\n",
      "New best: 1.7680454220680986e-06\n",
      "Epoch: 1000, CNMP Losses: -3.3476185099387075\n",
      "New best: 1.5091232796748955e-07\n",
      "Epoch: 2000, CNMP Losses: -4.827983657218516\n",
      "Epoch: 3000, CNMP Losses: -5.124947947952896\n",
      "New best: 2.511561802975848e-08\n",
      "Epoch: 4000, CNMP Losses: -5.255174568947405\n",
      "Epoch: 5000, CNMP Losses: -5.269271998725832\n",
      "Epoch: 6000, CNMP Losses: -5.27016359821707\n",
      "Epoch: 7000, CNMP Losses: -5.506937211524695\n",
      "Epoch: 8000, CNMP Losses: -5.487555700128898\n",
      "New best: 1.67121338989773e-08\n",
      "Epoch: 9000, CNMP Losses: -5.566774876767769\n",
      "Epoch: 10000, CNMP Losses: -5.54510827652365\n",
      "Epoch: 11000, CNMP Losses: -5.608870294798166\n",
      "Epoch: 12000, CNMP Losses: -5.6588899150136855\n",
      "Epoch: 13000, CNMP Losses: -5.706392304131761\n",
      "New best: 9.019631264095551e-09\n",
      "Epoch: 14000, CNMP Losses: -5.70954521086067\n",
      "Epoch: 15000, CNMP Losses: -5.740798598971218\n",
      "Epoch: 16000, CNMP Losses: -5.732462687995285\n",
      "Epoch: 17000, CNMP Losses: -5.724201388955116\n",
      "Epoch: 18000, CNMP Losses: -5.772347396967932\n",
      "Epoch: 19000, CNMP Losses: -5.78380740172416\n",
      "Epoch: 20000, CNMP Losses: -5.79161198605597\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "Cell \u001b[0;32mIn[5], line 41\u001b[0m\n\u001b[1;32m     38\u001b[0m traj_ids \u001b[38;5;241m=\u001b[39m torch\u001b[38;5;241m.\u001b[39mrandperm(x\u001b[38;5;241m.\u001b[39mshape[\u001b[38;5;241m0\u001b[39m])[:batch_size\u001b[38;5;241m*\u001b[39mepoch_iter]\u001b[38;5;241m.\u001b[39mchunk(epoch_iter)  \u001b[38;5;66;03m# [:batch_size*epoch_iter] because nof_trajectories may be indivisible by batch_size\u001b[39;00m\n\u001b[1;32m     40\u001b[0m \u001b[38;5;28;01mfor\u001b[39;00m i \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mrange\u001b[39m(epoch_iter):\n\u001b[0;32m---> 41\u001b[0m     \u001b[43mprepare_masked_batch\u001b[49m\u001b[43m(\u001b[49m\u001b[43mx\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mtraj_ids\u001b[49m\u001b[43m[\u001b[49m\u001b[43mi\u001b[49m\u001b[43m]\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m     43\u001b[0m     optimizer\u001b[38;5;241m.\u001b[39mzero_grad()\n\u001b[1;32m     44\u001b[0m     pred \u001b[38;5;241m=\u001b[39m model(obs, tar_x, obs_mask)\n",
      "Cell \u001b[0;32mIn[4], line 24\u001b[0m, in \u001b[0;36mprepare_masked_batch\u001b[0;34m(t, traj_ids)\u001b[0m\n\u001b[1;32m     21\u001b[0m n_ids \u001b[38;5;241m=\u001b[39m permuted_ids[:n]\n\u001b[1;32m     22\u001b[0m m_ids \u001b[38;5;241m=\u001b[39m permuted_ids[n:n\u001b[38;5;241m+\u001b[39mm]\n\u001b[0;32m---> 24\u001b[0m obs[i, :n, :dx] \u001b[38;5;241m=\u001b[39m \u001b[43mtraj\u001b[49m\u001b[43m[\u001b[49m\u001b[43mn_ids\u001b[49m\u001b[43m]\u001b[49m\n\u001b[1;32m     25\u001b[0m obs[i, :n, dx:] \u001b[38;5;241m=\u001b[39m (n_ids\u001b[38;5;241m/\u001b[39mt_steps)\u001b[38;5;241m.\u001b[39munsqueeze(\u001b[38;5;241m1\u001b[39m)\n\u001b[1;32m     26\u001b[0m obs_mask[i, :n] \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;01mTrue\u001b[39;00m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "import time\n",
    "import os\n",
    "\n",
    "timestamp = int(time.time())\n",
    "root_folder = f'outputs/tests/cnmp_basic/uni_sine/{str(timestamp)}/'\n",
    "\n",
    "if not os.path.exists(root_folder):\n",
    "    os.makedirs(root_folder)\n",
    "\n",
    "if not os.path.exists(f'{root_folder}saved_models/'):\n",
    "    os.makedirs(f'{root_folder}saved_models/')\n",
    "\n",
    "if not os.path.exists(f'{root_folder}img/'):\n",
    "    os.makedirs(f'{root_folder}img/')\n",
    "\n",
    "torch.save(y, f'{root_folder}y.pt')\n",
    "\n",
    "\n",
    "epochs = 500_000\n",
    "epoch_iter = num_demos//batch_size  # number of batches per epoch (e.g. 100//32 = 3)\n",
    "v_epoch_iter = num_val//batch_size  # number of batches per validation (e.g. 100//32 = 3)\n",
    "avg_loss = 0\n",
    "\n",
    "val_per_epoch = 1000\n",
    "min_vl = 1000000\n",
    "\n",
    "mse_loss = torch.nn.MSELoss()\n",
    "\n",
    "tl = []\n",
    "ve = []\n",
    "\n",
    "cnmp_tl_path = f'{root_folder}cnmp_training_loss.pt'\n",
    "cnmp_ve_path = f'{root_folder}cnmp_validation_error.pt'\n",
    "\n",
    "for epoch in range(epochs):\n",
    "    epoch_loss = 0\n",
    "\n",
    "    traj_ids = torch.randperm(x.shape[0])[:batch_size*epoch_iter].chunk(epoch_iter)  # [:batch_size*epoch_iter] because nof_trajectories may be indivisible by batch_size\n",
    "\n",
    "    for i in range(epoch_iter):\n",
    "        prepare_masked_batch(x, traj_ids[i])\n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        pred = model(obs, tar_x, obs_mask)\n",
    "        loss = model.loss(pred, tar_y, tar_mask)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        epoch_loss += loss.item()\n",
    "\n",
    "    epoch_loss = epoch_loss/epoch_iter\n",
    "    tl.append(epoch_loss)\n",
    "\n",
    "    if epoch % val_per_epoch == 0:\n",
    "        with torch.no_grad():\n",
    "            v_traj_ids = torch.randperm(vx.shape[0])[:batch_size*v_epoch_iter].chunk(v_epoch_iter)\n",
    "            val_err = 0\n",
    "\n",
    "            for j in range(v_epoch_iter):\n",
    "                prepare_masked_val_batch(vx, v_traj_ids[j])\n",
    "\n",
    "                p = model.val(val_obs, val_tar_x, val_obs_mask)\n",
    "                vp_means = p[:, :, :dy]\n",
    "                val_err += mse_loss(vp_means, val_tar_y).item()\n",
    "\n",
    "            val_err = val_err/v_epoch_iter\n",
    "\n",
    "            if val_err < min_vl:\n",
    "                min_vl = val_err\n",
    "                print(f'New best: {min_vl}')\n",
    "                torch.save(model.state_dict(), f'{root_folder}saved_models/cnmp.pt')\n",
    "\n",
    "            ve.append(val_err)\n",
    "\n",
    "    avg_loss += epoch_loss\n",
    "\n",
    "    if epoch % val_per_epoch == 0:\n",
    "        print(\"Epoch: {}, CNMP Losses: {}\".format(epoch, avg_loss/val_per_epoch))\n",
    "        avg_loss = 0\n",
    "\n",
    "torch.save(torch.Tensor(tl), cnmp_tl_path)\n",
    "torch.save(torch.Tensor(ve), cnmp_ve_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# from matplotlib.lines import Line2D\n",
    "\n",
    "# def get_validation_batch(o_ids=[15]):\n",
    "#     obs = torch.cat((vx[:, o_ids, :], vy[:, o_ids, :]), dim=-1)\n",
    "#     tar = vx[:, torch.arange(t_steps)]\n",
    "#     tar_val= vy[:, torch.arange(t_steps)]\n",
    "\n",
    "#     return obs, tar, tar_val\n",
    "\n",
    "# file_name = '1691573215'\n",
    "\n",
    "# # Testing the best model\n",
    "# model = CNP(input_dim=1, hidden_dim=204, output_dim=1, n_max_obs=6, n_max_tar=6, num_layers=3, batch_size=batch_size)\n",
    "# model.load_state_dict(torch.load(f'saved_models/cnp_synth_{file_name}.pt'))\n",
    "# model.eval()\n",
    "\n",
    "# ids = [15]\n",
    "# o, t, tr = get_validation_batch(ids)\n",
    "# o, t, tr = o.cpu(), t.cpu(), tr.cpu()\n",
    "\n",
    "# with torch.no_grad():\n",
    "#     pred, _ = model(o, t)\n",
    "\n",
    "# p = pred.cpu().numpy()\n",
    "# t, tr = t.cpu().numpy(), tr.cpu().numpy()\n",
    "\n",
    "# for i in range(vx.shape[0]):\n",
    "#     plt.plot(range(t_steps), vy[i, :, 0].cpu(), 'k', alpha=0.15)\n",
    "\n",
    "# for i in range(len(ids)):\n",
    "#     plt.plot([ids[i]]*t_steps, torch.linspace(0, 1, t_steps), 'k--', alpha=0.75)\n",
    "\n",
    "# for i in range(4):\n",
    "#     plt.plot(range(t_steps), p[i, :, 0], 'b', alpha=0.6)\n",
    "\n",
    "# line0 = Line2D([0], [0], label=f'ground truth', color='k', alpha=0.15)\n",
    "# line1 = Line2D([0], [0], label=f'prediction', color='b', alpha=0.6)\n",
    "# line2 = Line2D([0], [0], label=f'observation', color='k', linestyle='--', alpha=0.75)\n",
    "\n",
    "# handles = [line0, line1, line2]\n",
    "\n",
    "# plt.legend(handles=handles, loc='best')\n",
    "# plt.title(f'CNP (Easy)')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# ids = [60]\n",
    "# o, t, tr = get_validation_batch(ids)\n",
    "# o, t, tr = o.cpu(), t.cpu(), tr.cpu()\n",
    "\n",
    "# with torch.no_grad():\n",
    "#     pred, _ = model(o, t)\n",
    "\n",
    "# p = pred.cpu().numpy()\n",
    "# t, tr = t.cpu().numpy(), tr.cpu().numpy()\n",
    "\n",
    "# for i in range(vx.shape[0]):\n",
    "#     plt.plot(range(t_steps), vy[i, :, 0].cpu(), 'k', alpha=0.15)\n",
    "\n",
    "# for i in range(len(ids)):\n",
    "#     plt.plot([ids[i]]*t_steps, torch.linspace(0, 1, t_steps), 'k--', alpha=0.75)\n",
    "\n",
    "# for i in range(4):\n",
    "#     plt.plot(range(t_steps), p[i, :, 0], 'b', alpha=0.6)\n",
    "\n",
    "# line0 = Line2D([0], [0], label=f'ground truth', color='k', alpha=0.15)\n",
    "# line1 = Line2D([0], [0], label=f'prediction', color='b', alpha=0.6)\n",
    "# line2 = Line2D([0], [0], label=f'observation', color='k', linestyle='--', alpha=0.75)\n",
    "\n",
    "# handles = [line0, line1, line2]\n",
    "\n",
    "# plt.legend(handles=handles, loc='best')\n",
    "# plt.title(f'CNP (Difficult)')"
   ]
  }
 ],
 "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": "3.8.10"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}