{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import torch\n",
    "\n",
    "torch.cuda.is_available()\n",
    "os.environ[\"WANDB_ENABLED\"] = \"false\"\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"-1\"\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "device = torch.device(f\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\n",
    "from models.Tiffusion import tiffusion\n",
    "# from models.CSDI import tiffusion\n",
    "\n",
    "model = tiffusion.Tiffusion(\n",
    "    seq_length=365,\n",
    "    feature_size=3,\n",
    "    n_layer_enc=6,\n",
    "    n_layer_dec=4,\n",
    "    d_model=128,\n",
    "    timesteps=500,\n",
    "    sampling_timesteps=200,\n",
    "    loss_type='l1',\n",
    "    beta_schedule='cosine',\n",
    "    n_heads=8,\n",
    "    mlp_hidden_times=4,\n",
    "    attn_pd=0.0,\n",
    "    resid_pd=0.0,\n",
    "    kernel_size=1,\n",
    "    padding_size=0,\n",
    "    control_signal=[]\n",
    ").to(device)\n",
    "\n",
    "model.load_state_dict(torch.load(\"./weight/checkpoint-10.pt\", map_location='cpu', weights_only=True)[\"model\"])\n",
    "# model.load_state_dict(torch.load(\"../../../data/CSDI/ckpt_baseline_365/checkpoint-10.pt\", map_location='cpu', weights_only=True)[\"model\"])\n",
    "\n",
    "\n",
    "coef = 1.0e-2\n",
    "stepsize =  5.0e-2\n",
    "sampling_steps = 100 # 这个可以调整 100-500都行 快慢和准度 tradeoff\n",
    "seq_length = 365\n",
    "feature_dim = 3\n",
    "print(f\"seq_length: {seq_length}, feature_dim: {feature_dim}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Sampling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "anchor_value = [\n",
    "    # (time, feature_id, y-value, confidence)\n",
    "    (0, 0, 0.04, 1.0),\n",
    "    (2, 0, 0.58, 1.0),\n",
    "    # (6, 0, 0.27, 0.5),\n",
    "    # (10, 0, 0.04, 1.0),\n",
    "    # (12, 0, 0.58, 0.001),\n",
    "    # (16, 0, 0.27, 0.5),\n",
    "    # (20, 0, 0.04, 1.0),\n",
    "    # (22, 0, 0.58, 0.001),\n",
    "    # (26, 0, 0.27, 0.5),\n",
    "    # (30, 0, 0.04, 1.0),\n",
    "    # (32, 0, 0.58, 0.001),\n",
    "    # (36, 0, 0.27, 0.5),\n",
    "    # (40, 0, 0.04, 1.0),\n",
    "    # (42, 0, 0.58, 0.001),\n",
    "    # (46, 0, 0.27, 0.5),\n",
    "    # (50, 0, 0.04, 1.0),\n",
    "    # (52, 0, 0.58, 0.001),\n",
    "    # (56, 0, 0.27, 0.5),\n",
    "    # (60, 0, 0.04, 1.0),\n",
    "    # (62, 0, 0.58, 0.001),\n",
    "    # (66, 0, 0.27, 0.5),\n",
    "]\n",
    "\n",
    "observed_points = torch.zeros((seq_length, feature_dim)).to(device)\n",
    "observed_mask = torch.zeros((seq_length, feature_dim)).to(device)\n",
    "\n",
    "for time, feature_id, y_value, confidence in anchor_value:\n",
    "    observed_points[time, feature_id] = y_value\n",
    "    observed_mask[time, feature_id] = confidence\n",
    "\n",
    "auc = -10\n",
    "auc_weight = 10.0\n",
    "with torch.no_grad():\n",
    "    results = model.predict_weighted_points(\n",
    "        observed_points, # (seq_length, feature_dim)\n",
    "        observed_mask, # (seq_length, feature_dim)\n",
    "        coef,  # fixed\n",
    "        stepsize, # fixed\n",
    "        sampling_steps, # fixed\n",
    "        # model_control_signal=model_control_signal,\n",
    "        gradient_control_signal={\n",
    "            \"auc\": auc, \"auc_weight\": auc_weight,\n",
    "        },\n",
    "    )\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "results.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "plt.plot(results[:,0], label=\"Predicted Feature 0\")\n",
    "for time, feature_id, y_value, confidence in anchor_value:\n",
    "    plt.scatter(time, y_value, c='r')\n",
    "plt.show()\n",
    "plt.plot(results[:,1], label=\"Predicted Feature 1\")\n",
    "plt.show()\n",
    "plt.plot(results[:,2], label=\"Predicted Feature 2\")\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "rag",
   "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": "3.10.14"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}