{ "cells": [ { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "import pickle\n", "import pycuda.driver as cuda\n", "import sys\n", "import os\n", "from datetime import datetime\n", "from stable_baselines3 import PPO\n", "\n", "current_dir = os.path.dirname(os.path.abspath(\"__file__\"))\n", "parent_dir = os.path.abspath(os.path.join(current_dir, os.pardir))\n", "output_dir = os.path.join(parent_dir, \"output\")\n", "os.makedirs(output_dir, exist_ok=True)\n", "sys.path.append(parent_dir)\n", "\n", "cuda.init()\n", "context2 = cuda.Device(2).make_context()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "with open(os.path.join(output_dir, \"d1a3o12_re100.pkl\"), 'rb') as f:\n", " data = pickle.load(f)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from gym_env import CustomEnv\n", "context2.push()\n", "vec_env = CustomEnv(device_id=2)\n", "context2.pop()\n", "model = PPO.load(os.path.join(parent_dir, \"models\", \"d1a3o12_re100.zip\"), env=vec_env, device=\"cuda:2\")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "context2.push()\n", "model_env = model.get_env()\n", "# obs = model_env.reset()\n", "obs_init = np.array(model_env.envs[0].save_states.copy())\n", "context2.pop()\n", "\n", "obs_init[:, 6:12] = obs_init[:, 6:12]/model_env.envs[0].force_norm_fact.copy()\n", "sens_deviation = model_env.envs[0].sens_deviation.copy()\n", "sens_norm_fact = model_env.envs[0].sens_norm_fact.copy()\n", "for i in range(6):\n", " obs_init[:, i] = (obs_init[:, i] - sens_deviation[i]) / sens_norm_fact[i]" ] }, { "cell_type": "code", "execution_count": 15, "metadata": {}, "outputs": [], "source": [ "import pandas as pd\n", "from darts import TimeSeries\n", "from darts.models import RNNModel # 或者 NBEATSModel、TFTModel 等\n", "from darts.models import TFTModel\n", "from darts.dataprocessing.transformers import Scaler\n", "from typing import List" ] }, { "cell_type": "code", "execution_count": 18, "metadata": {}, "outputs": [], "source": [ "set_obs_train = []\n", "set_act_train = []\n", "index = pd.RangeIndex(start=0, stop=480, step=1)\n", "for i in range(len(data)):\n", " obs_episolde = np.array(data[i]['observations'])[:, 0]\n", " if np.any(np.all(obs_episolde == 0, axis=1)):\n", " # print(f\"Episode {i} contains a row with all zeros in observations.\")\n", " continue\n", " else:\n", " obs_full = np.concatenate((obs_init, obs_episolde), axis=0)\n", " act_init = np.zeros((len(obs_init), 3))\n", " act_episolde = np.array(data[i]['actions'])\n", " act_full = np.concatenate((act_init, act_episolde), axis=0)\n", "\n", " obs_ts = TimeSeries.from_times_and_values(\n", " times=index,\n", " values=obs_full.astype(np.float32), \n", " columns=[f\"obs{dim+1}\" for dim in range(12)]\n", " )\n", " act_ts = TimeSeries.from_times_and_values(\n", " times=index,\n", " values=act_full.astype(np.float32), \n", " columns=[f\"act{dim+1}\" for dim in range(3)]\n", " )\n", "\n", " set_obs_train.append(obs_ts)\n", " set_act_train.append(act_ts)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "Trainer will use only 1 of 4 GPUs because it is running inside an interactive / notebook environment. You may try to set `Trainer(devices=4)` but please note that multi-GPU inside interactive / notebook environments is considered experimental and unstable. Your mileage may vary.\n", "GPU available: True (cuda), used: True\n", "TPU available: False, using: 0 TPU cores\n", "HPU available: False, using: 0 HPUs\n", "LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0,1,2,3]\n" ] }, { "name": "stderr", "output_type": "stream", "text": [ "\n", " | Name | Type | Params | Mode \n", "------------------------------------------------------------------------------------------------\n", "0 | train_metrics | MetricCollection | 0 | train\n", "1 | val_metrics | MetricCollection | 0 | train\n", "2 | input_embeddings | _MultiEmbedding | 0 | train\n", "3 | static_covariates_vsn | _VariableSelectionNetwork | 0 | train\n", "4 | encoder_vsn | _VariableSelectionNetwork | 17.8 K | train\n", "5 | decoder_vsn | _VariableSelectionNetwork | 896 | train\n", "6 | static_context_grn | _GatedResidualNetwork | 4.3 K | train\n", "7 | static_context_hidden_encoder_grn | _GatedResidualNetwork | 4.3 K | train\n", "8 | static_context_cell_encoder_grn | _GatedResidualNetwork | 4.3 K | train\n", "9 | static_context_enrichment | _GatedResidualNetwork | 4.3 K | train\n", "10 | lstm_encoder | LSTM | 8.4 K | train\n", "11 | lstm_decoder | LSTM | 8.4 K | train\n", "12 | post_lstm_gan | _GateAddNorm | 2.2 K | train\n", "13 | static_enrichment_grn | _GatedResidualNetwork | 5.3 K | train\n", "14 | multihead_attn | _InterpretableMultiHeadAttention | 2.6 K | train\n", "15 | post_attn_gan | _GateAddNorm | 2.2 K | train\n", "16 | feed_forward_block | _GatedResidualNetwork | 4.3 K | train\n", "17 | pre_output_gan | _GateAddNorm | 2.2 K | train\n", "18 | output_layer | Linear | 6.7 K | train\n", "------------------------------------------------------------------------------------------------\n", "78.2 K Trainable params\n", "0 Non-trainable params\n", "78.2 K Total params\n", "0.313 Total estimated model params size (MB)\n", "382 Modules in train mode\n", "0 Modules in eval mode\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Epoch 49: 100%|██████████| 12/12 [00:00<00:00, 21.99it/s, train_loss=0.133]" ] }, { "name": "stderr", "output_type": "stream", "text": [ "`Trainer.fit` stopped: `max_epochs=50` reached.\n" ] }, { "name": "stdout", "output_type": "stream", "text": [ "Epoch 49: 100%|██████████| 12/12 [00:00<00:00, 21.93it/s, train_loss=0.133]\n" ] }, { "data": { "text/plain": [ "TFTModel(output_chunk_shift=0, hidden_size=32, lstm_layers=1, num_attention_heads=4, full_attention=False, feed_forward=GatedResidualNetwork, dropout=0.1, hidden_continuous_size=8, categorical_embedding_sizes=None, add_relative_index=True, loss_fn=None, likelihood=None, norm_type=LayerNorm, use_static_covariates=True, input_chunk_length=120, output_chunk_length=1, batch_size=32, n_epochs=50, random_state=42)" ] }, "execution_count": 19, "metadata": {}, "output_type": "execute_result" } ], "source": [ "model_tft = TFTModel(\n", " input_chunk_length=120,\n", " output_chunk_length=1, # 预测1步\n", " hidden_size=32,\n", " lstm_layers=1,\n", " dropout=0.1,\n", " batch_size=32,\n", " n_epochs=50,\n", " add_relative_index=True,\n", " random_state=42,\n", " # 注意：TFT支持past_covariates与future_covariates，也可设置transformer头等更多超参\n", ")\n", "\n", "model_tft.fit(\n", " series=obs_ts,\n", " past_covariates=act_ts\n", ")" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "for i in range(360):" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "plot_data = obs_full[:, 11]\n", "import matplotlib.pyplot as plt\n", "plt.plot(plot_data)" ] } ], "metadata": { "kernelspec": { "display_name": "pycuda_3_10", "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": "3.10.13" } }, "nbformat": 4, "nbformat_minor": 2 }