fmda/fmda_rnn_serial.ipynb

   1 {
   2  "cells": [
   3   {
   4    "cell_type": "markdown",
   5    "id": "83b774b3-ef55-480a-b999-506676e49145",
   6    "metadata": {},
   7    "source": [
   8     "# v2.1 run RNN strategy serial by Location\n",
   9     "\n",
  10     "This version of the RNN runs the model on each location separately, one at a time. Two main runs:\n",
  11     "1. Run separate model at each location - training and prediction at least location independently - training mode periods 0:train_ind (was 0:h2), then prediction in test_ind:end. Validation data, if any, are from train_ind:test_ind\n",
  12     "2. Run same model with multiple fitting calls 0:train_ind at different locations, compare prediction accuracy in test_ind:end  at for all location. \n"
  13    ]
  14   },
  15   {
  16    "cell_type": "code",
  17    "execution_count": null,
  18    "id": "83cc1dc4-3dcb-4325-9263-58101a3dc378",
  19    "metadata": {},
  20    "outputs": [],
  21    "source": [
  22     "import numpy as np\n",
  23     "from utils import print_dict_summary, print_first, str2time, logging_setup\n",
  24     "import pickle\n",
  25     "import logging\n",
  26     "import os.path as osp\n",
  27     "from moisture_rnn_pkl import pkl2train\n",
  28     "from moisture_rnn import RNNParams, RNNData, RNN \n",
  29     "from utils import hash2, read_yml, read_pkl, retrieve_url\n",
  30     "from moisture_rnn import RNN\n",
  31     "import reproducibility\n",
  32     "from data_funcs import rmse, to_json\n",
  33     "from moisture_models import run_augmented_kf\n",
  34     "import copy\n",
  35     "import pandas as pd\n",
  36     "import matplotlib.pyplot as plt\n",
  37     "import yaml"
  38    ]
  39   },
  40   {
  41    "cell_type": "code",
  42    "execution_count": null,
  43    "id": "17db9b90-a931-4674-a447-5b8ffbcdc86a",
  44    "metadata": {},
  45    "outputs": [],
  46    "source": [
  47     "logging_setup()"
  48    ]
  49   },
  50   {
  51    "cell_type": "code",
  52    "execution_count": null,
  53    "id": "35319c1c-7849-4b8c-8262-f5aa6656e0c7",
  54    "metadata": {},
  55    "outputs": [],
  56    "source": [
  57     "retrieve_url(\n",
  58     "    url = \"https://demo.openwfm.org/web/data/fmda/dicts/test_CA_202401.pkl\", \n",
  59     "    dest_path = \"data/test_CA_202401.pkl\")"
  60    ]
  61   },
  62   {
  63    "cell_type": "code",
  64    "execution_count": null,
  65    "id": "eabdbd9c-07d9-4bae-9851-cca79f321895",
  66    "metadata": {},
  67    "outputs": [],
  68    "source": [
  69     "repro_file = \"data/reproducibility_dict_v2_TEST.pkl\"\n",
  70     "file_names=['fmda_nw_202401-05_f05.pkl']\n",
  71     "file_dir='data'\n",
  72     "file_paths = [osp.join(file_dir,file_name) for file_name in file_names]"
  73    ]
  74   },
  75   {
  76    "cell_type": "code",
  77    "execution_count": null,
  78    "id": "dcca6185-e799-4dd1-8acb-87ad33c411d7",
  79    "metadata": {},
  80    "outputs": [],
  81    "source": [
  82     "# read/write control\n",
  83     "train_file='train.pkl'\n",
  84     "train_create=False   # if false, read\n",
  85     "train_write=False\n",
  86     "train_read=True"
  87    ]
  88   },
  89   {
  90    "cell_type": "code",
  91    "execution_count": null,
  92    "id": "bc0a775b-b587-42ef-8576-e36dc0be3a75",
  93    "metadata": {
  94     "scrolled": true
  95    },
  96    "outputs": [],
  97    "source": [
  98     "repro = read_pkl(repro_file)\n",
  99     "\n",
 100     "if train_create:\n",
 101     "    logging.info('creating the training cases from files %s',file_paths)\n",
 102     "    # osp.join works on windows too, joins paths using \\ or /\n",
 103     "    train = pkl2train(file_paths)\n",
 104     "if train_write:\n",
 105     "    with open(train_file, 'wb') as file:\n",
 106     "        logging.info('Writing the rain cases into file %s',train_file)\n",
 107     "        pickle.dump(train, file)\n",
 108     "if train_read:\n",
 109     "    logging.info('Reading the train cases from file %s',train_file)\n",
 110     "    train = read_pkl(train_file)"
 111    ]
 112   },
 113   {
 114    "cell_type": "code",
 115    "execution_count": null,
 116    "id": "211a1c2f-ba8d-40b8-b29c-daa38af97a26",
 117    "metadata": {},
 118    "outputs": [],
 119    "source": [
 120     "params_all = read_yml(\"params.yaml\")\n",
 121     "print(params_all.keys())"
 122    ]
 123   },
 124   {
 125    "cell_type": "code",
 126    "execution_count": null,
 127    "id": "698df86b-8550-4135-81df-45dbf503dd4e",
 128    "metadata": {},
 129    "outputs": [],
 130    "source": [
 131     "# from module_param_sets import param_sets"
 132    ]
 133   },
 134   {
 135    "cell_type": "code",
 136    "execution_count": null,
 137    "id": "4b0c9a9b-dd02-4251-aa4a-2acc1101e153",
 138    "metadata": {},
 139    "outputs": [],
 140    "source": [
 141     "param_sets_keys=['rnn']\n",
 142     "cases=list(train.keys())[0:10]\n",
 143     "# cases=list(train.keys())\n",
 144     "# cases.remove('reproducibility')\n",
 145     "cases"
 146    ]
 147   },
 148   {
 149    "cell_type": "code",
 150    "execution_count": null,
 151    "id": "dd22baf2-59d2-460e-8c47-b20116dd5982",
 152    "metadata": {},
 153    "outputs": [],
 154    "source": [
 155     "logging.info('Running over parameter sets %s',param_sets_keys)\n",
 156     "logging.info('Running over cases %s',cases)"
 157    ]
 158   },
 159   {
 160    "cell_type": "markdown",
 161    "id": "802f3eef-1702-4478-b6e3-2288a6edae24",
 162    "metadata": {},
 163    "source": [
 164     "## Run Reproducibility Case"
 165    ]
 166   },
 167   {
 168    "cell_type": "code",
 169    "execution_count": null,
 170    "id": "69a3adb9-39fd-4c0c-9c9b-aaa2a9a3af40",
 171    "metadata": {},
 172    "outputs": [],
 173    "source": [
 174     "params = repro['repro_info']['params']\n",
 175     "print(type(params))\n",
 176     "print(params)\n",
 177     "\n",
 178     "# Set up input data\n",
 179     "rnn_dat = RNNData(repro, scaler = params['scaler'], features_list = params['features_list'])\n",
 180     "rnn_dat.train_test_split(\n",
 181     "    train_frac = params['train_frac'],\n",
 182     "    val_frac = params['val_frac']\n",
 183     ")\n",
 184     "rnn_dat.scale_data()\n",
 185     "rnn_dat.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])"
 186    ]
 187   },
 188   {
 189    "cell_type": "code",
 190    "execution_count": null,
 191    "id": "855703c4-d7a9-4579-bca7-7c737a81d0de",
 192    "metadata": {},
 193    "outputs": [],
 194    "source": [
 195     "reproducibility.set_seed(123)\n",
 196     "rnn = RNN(params)\n",
 197     "m, errs = rnn.run_model(rnn_dat, reproducibility_run=True)"
 198    ]
 199   },
 200   {
 201    "cell_type": "markdown",
 202    "id": "49e31fdd-4c14-4a81-9e2b-4c6ba94d1f83",
 203    "metadata": {},
 204    "source": [
 205     "## Separate Models by Location"
 206    ]
 207   },
 208   {
 209    "cell_type": "code",
 210    "execution_count": null,
 211    "id": "e11e7c83-183f-48ba-abd8-a6aedff66090",
 212    "metadata": {},
 213    "outputs": [],
 214    "source": [
 215     "# Set up output dictionaries\n",
 216     "outputs_kf = {}\n",
 217     "outputs_rnn = {}"
 218    ]
 219   },
 220   {
 221    "cell_type": "code",
 222    "execution_count": null,
 223    "id": "dc5b47bd-4fbc-44b8-b2dd-d118e068b450",
 224    "metadata": {
 225     "scrolled": true
 226    },
 227    "outputs": [],
 228    "source": [
 229     "\n",
 230     "for k in param_sets_keys:\n",
 231     "    params = RNNParams(params_all[k])\n",
 232     "    print(\"~\"*80)\n",
 233     "    print(\"Running with params:\")\n",
 234     "    print(params)\n",
 235     "    # Increase Val Frac so no errors, TODO fix validation\n",
 236     "    params.update({\n",
 237     "        'train_frac': .9,\n",
 238     "        'val_frac': .05,\n",
 239     "        'activation': ['relu', 'relu'],\n",
 240     "        'epochs': 10,\n",
 241     "        'dense_units': 10,\n",
 242     "        'rnn_layers': 2       \n",
 243     "    })\n",
 244     "    for case in cases:\n",
 245     "        print(\"~\"*50)\n",
 246     "        logging.info('Processing case %s',case)\n",
 247     "        print_dict_summary(train[case])\n",
 248     "        # Format data & Run Model\n",
 249     "        # rnn_dat = create_rnn_data2(train[case], params)\n",
 250     "        rnn_dat = RNNData(train[case], scaler = params['scaler'], features_list = params['features_list'])\n",
 251     "        rnn_dat.train_test_split(\n",
 252     "            train_frac = params['train_frac'],\n",
 253     "            val_frac = params['val_frac']\n",
 254     "        )\n",
 255     "        rnn_dat.scale_data()\n",
 256     "        rnn_dat.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])\n",
 257     "        reproducibility.set_seed()\n",
 258     "        rnn = RNN(params)\n",
 259     "        m, errs = rnn.run_model(rnn_dat)\n",
 260     "        # Add model output to case\n",
 261     "        train[case]['m']=m\n",
 262     "        # Get RMSE Prediction Error\n",
 263     "        print(f\"RMSE: {errs}\")\n",
 264     "        outputs_rnn[case] = {'case':case, 'errs': errs.copy()}\n",
 265     "        \n",
 266     "        # Run Augmented KF\n",
 267     "        print('Running Augmented KF')\n",
 268     "        train[case]['h2'] = train[case]['hours'] // 2\n",
 269     "        train[case]['scale_fm'] = 1\n",
 270     "        m, Ec = run_augmented_kf(train[case])\n",
 271     "        m = m*rnn_dat['scale_fm']\n",
 272     "        y = rnn_dat['y']*rnn_dat['scale_fm']          \n",
 273     "        train[case]['m'] = m\n",
 274     "        print(f\"KF RMSE: {rmse(m,y)}\")\n",
 275     "        outputs_kf[case] = {'case':case, 'errs': rmse(m,y)}\n",
 276     "\n",
 277     "        # Save Outputs \n",
 278     "        to_json(outputs_rnn, \"rnn_errs.json\")\n",
 279     "        to_json(outputs_kf, \"kf_errs.json\")"
 280    ]
 281   },
 282   {
 283    "cell_type": "code",
 284    "execution_count": null,
 285    "id": "15384e4d-b8ec-4700-bdc2-83b0433d11c9",
 286    "metadata": {},
 287    "outputs": [],
 288    "source": [
 289     "logging.info('fmda_rnn_serial.ipynb done')"
 290    ]
 291   },
 292   {
 293    "cell_type": "code",
 294    "execution_count": null,
 295    "id": "d0e78fb3-b501-49d6-81a9-1a13da0134a0",
 296    "metadata": {},
 297    "outputs": [],
 298    "source": [
 299     "import importlib\n",
 300     "import moisture_rnn\n",
 301     "importlib.reload(moisture_rnn)\n",
 302     "from moisture_rnn import RNN"
 303    ]
 304   },
 305   {
 306    "cell_type": "code",
 307    "execution_count": null,
 308    "id": "37053436-8dfe-4c40-8614-811817e83782",
 309    "metadata": {},
 310    "outputs": [],
 311    "source": [
 312     "for k in outputs_rnn:\n",
 313     "    print(\"~\"*50)\n",
 314     "    print(outputs_rnn[k]['case'])\n",
 315     "    print(outputs_rnn[k]['errs']['prediction'])"
 316    ]
 317   },
 318   {
 319    "cell_type": "code",
 320    "execution_count": null,
 321    "id": "9154d5f7-015f-4ef7-af45-020410a1ea65",
 322    "metadata": {},
 323    "outputs": [],
 324    "source": [
 325     "for k in outputs_kf:\n",
 326     "    print(\"~\"*50)\n",
 327     "    print(outputs_kf[k]['case'])\n",
 328     "    print(outputs_kf[k]['errs'])"
 329    ]
 330   },
 331   {
 332    "cell_type": "markdown",
 333    "id": "f3c1c299-1655-4c64-a458-c7723db6ea6d",
 334    "metadata": {},
 335    "source": [
 336     "### TODO: FIX SCALING in Scheme below\n",
 337     "\n",
 338     "Scaling is done separately in each now."
 339    ]
 340   },
 341   {
 342    "cell_type": "markdown",
 343    "id": "0c0c3470-30f5-4915-98a7-dcdf5760d482",
 344    "metadata": {},
 345    "source": [
 346     "## Training at Multiple Locations\n",
 347     "\n",
 348     "Still sequential"
 349    ]
 350   },
 351   {
 352    "cell_type": "code",
 353    "execution_count": null,
 354    "id": "dd1aca73-7279-473e-b2a3-95aa1db7b1a8",
 355    "metadata": {},
 356    "outputs": [],
 357    "source": [
 358     "params = RNNParams(params_all['rnn'])\n",
 359     "params.update({\n",
 360     "    'epochs': 1, # less epochs since it is per location\n",
 361     "    'activation': ['relu', 'relu'],\n",
 362     "    'train_frac': .9,\n",
 363     "    'val_frac': .05,    \n",
 364     "    'dense_units': 10,\n",
 365     "    'rnn_layers': 2\n",
 366     "})\n",
 367     "\n",
 368     "# rnn_dat = create_rnn_data2(train[cases[0]], params)\n",
 369     "rnn_dat = RNNData(train[cases[0]], params['scaler'], params['features_list'])\n",
 370     "rnn_dat.train_test_split(\n",
 371     "    train_frac = params['train_frac'],\n",
 372     "    val_frac = params['val_frac']\n",
 373     ")\n",
 374     "rnn_dat.scale_data()\n",
 375     "rnn_dat.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])"
 376    ]
 377   },
 378   {
 379    "cell_type": "code",
 380    "execution_count": null,
 381    "id": "65b2f9a3-a8f2-4ac1-8e4d-ba38a86eaf40",
 382    "metadata": {},
 383    "outputs": [],
 384    "source": [
 385     "reproducibility.set_seed()\n",
 386     "rnn = RNN(params)"
 387    ]
 388   },
 389   {
 390    "cell_type": "code",
 391    "execution_count": null,
 392    "id": "47a85ef2-8145-4de8-9f2e-86622306ffd8",
 393    "metadata": {
 394     "scrolled": true
 395    },
 396    "outputs": [],
 397    "source": [
 398     "print(\"~\"*80)\n",
 399     "print(\"Running with params:\")\n",
 400     "print(params)\n",
 401     "\n",
 402     "for case in cases[0:10]:\n",
 403     "    print(\"~\"*50)\n",
 404     "    logging.info('Processing case %s',case)\n",
 405     "    print_dict_summary(train[case])\n",
 406     "    rnn_dat_temp = RNNData(train[case], params['scaler'], params['features_list'])\n",
 407     "    rnn_dat_temp.train_test_split(\n",
 408     "        train_frac = params['train_frac'],\n",
 409     "        val_frac = params['val_frac']\n",
 410     "    )\n",
 411     "    rnn_dat_temp.scale_data()\n",
 412     "    rnn_dat_temp.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])\n",
 413     "    rnn.fit(rnn_dat_temp['X_train'], rnn_dat_temp['y_train'],\n",
 414     "           validation_data=(rnn_dat_temp['X_val'], rnn_dat_temp['y_val']))\n",
 415     "    # run_rnn_pkl(train[case],param_sets[i])"
 416    ]
 417   },
 418   {
 419    "cell_type": "markdown",
 420    "id": "a0421b8d-49aa-4409-8cbf-7732f1137838",
 421    "metadata": {},
 422    "source": [
 423     "### Predict "
 424    ]
 425   },
 426   {
 427    "cell_type": "code",
 428    "execution_count": null,
 429    "id": "63d7854a-94f7-425c-9561-4fe518e044bb",
 430    "metadata": {
 431     "scrolled": true
 432    },
 433    "outputs": [],
 434    "source": [
 435     "# Predict Cases Used in Training\n",
 436     "rmses = []\n",
 437     "inds = np.arange(0,10)\n",
 438     "train_keys = list(train.keys())\n",
 439     "for i in inds:\n",
 440     "    print(\"~\"*50)\n",
 441     "    case = train_keys[i]\n",
 442     "    print(f\"Predicting case {case}\")\n",
 443     "    # rnn_dat = create_rnn_data2(train[case], params)\n",
 444     "    rnn_dat_temp = RNNData(train[case], params['scaler'], params['features_list'])\n",
 445     "    rnn_dat_temp.train_test_split(\n",
 446     "        train_frac = params['train_frac'],\n",
 447     "        val_frac = params['val_frac']\n",
 448     "    )\n",
 449     "    rnn_dat_temp.scale_data()\n",
 450     "    rnn_dat_temp.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])\n",
 451     "    X_temp = rnn_dat_temp.scale_all_X()\n",
 452     "    m = rnn.predict(X_temp)\n",
 453     "    test_ind = rnn_dat['test_ind']\n",
 454     "    rmses.append(rmse(m[test_ind:], rnn_dat['y_test'].flatten()))"
 455    ]
 456   },
 457   {
 458    "cell_type": "code",
 459    "execution_count": null,
 460    "id": "2a5423e0-778b-4f69-9ed0-f0082a1fefe5",
 461    "metadata": {},
 462    "outputs": [],
 463    "source": [
 464     "rmses"
 465    ]
 466   },
 467   {
 468    "cell_type": "code",
 469    "execution_count": null,
 470    "id": "45c9caae-7ced-4f21-aa05-c9b125e8fdcb",
 471    "metadata": {},
 472    "outputs": [],
 473    "source": [
 474     "pd.DataFrame({'Case': list(train.keys())[0:10], 'RMSE': rmses}).style.hide(axis=\"index\")"
 475    ]
 476   },
 477   {
 478    "cell_type": "code",
 479    "execution_count": null,
 480    "id": "f710f482-b600-4ea5-9a8a-823a13b4ec7a",
 481    "metadata": {
 482     "scrolled": true
 483    },
 484    "outputs": [],
 485    "source": [
 486     "# Predict New Locations\n",
 487     "rmses = []\n",
 488     "for i, case in enumerate(list(train.keys())[10:100]):\n",
 489     "    print(\"~\"*50)\n",
 490     "    print(f\"Predicting case {case}\")\n",
 491     "    rnn_dat_temp = RNNData(train[case], params['scaler'], params['features_list'])\n",
 492     "    rnn_dat_temp.train_test_split(\n",
 493     "        train_frac = params['train_frac'],\n",
 494     "        val_frac = params['val_frac']\n",
 495     "    )\n",
 496     "    rnn_dat_temp.scale_data()\n",
 497     "    rnn_dat_temp.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])\n",
 498     "    X = rnn_dat_temp.scale_all_X()\n",
 499     "    m = rnn.predict(X)\n",
 500     "    train[case]['m'] = m\n",
 501     "    test_ind = rnn_dat['test_ind']\n",
 502     "    rmses.append(rmse(m[test_ind:], rnn_dat.y_test.flatten()))\n",
 503     "\n",
 504     "df = pd.DataFrame({'Case': list(train.keys())[10:100], 'RMSE': rmses})"
 505    ]
 506   },
 507   {
 508    "cell_type": "code",
 509    "execution_count": null,
 510    "id": "d793ac87-d94b-4b16-a271-46cdc259b4fe",
 511    "metadata": {},
 512    "outputs": [],
 513    "source": [
 514     "df[0:5].style.hide(axis=\"index\")"
 515    ]
 516   },
 517   {
 518    "cell_type": "code",
 519    "execution_count": null,
 520    "id": "b99606d1-bd46-4041-8303-1bcbb196f6f4",
 521    "metadata": {},
 522    "outputs": [],
 523    "source": [
 524     "df"
 525    ]
 526   },
 527   {
 528    "cell_type": "code",
 529    "execution_count": null,
 530    "id": "52ec264d-d4b7-444c-b623-002d6383da30",
 531    "metadata": {},
 532    "outputs": [],
 533    "source": [
 534     "df.RMSE.mean()"
 535    ]
 536   },
 537   {
 538    "cell_type": "code",
 539    "execution_count": null,
 540    "id": "998922cd-46bb-4063-8284-0497e19c39b0",
 541    "metadata": {},
 542    "outputs": [],
 543    "source": [
 544     "plt.hist(df.RMSE)"
 545    ]
 546   },
 547   {
 548    "cell_type": "code",
 549    "execution_count": null,
 550    "id": "889f3bbb-9fb2-4621-9e93-1d0bc0f83e01",
 551    "metadata": {},
 552    "outputs": [],
 553    "source": []
 554   },
 555   {
 556    "cell_type": "code",
 557    "execution_count": null,
 558    "id": "fe407f61-15f2-4086-a386-7d7a5bb90d26",
 559    "metadata": {},
 560    "outputs": [],
 561    "source": []
 562   },
 563   {
 564    "cell_type": "code",
 565    "execution_count": null,
 566    "id": "2fdb63b3-68b8-4877-a7a2-f63257cb29d5",
 567    "metadata": {},
 568    "outputs": [],
 569    "source": []
 570   },
 571   {
 572    "cell_type": "code",
 573    "execution_count": null,
 574    "id": "5c7563c5-a880-45c7-8381-8ce4e1a44216",
 575    "metadata": {},
 576    "outputs": [],
 577    "source": []
 578   },
 579   {
 580    "cell_type": "code",
 581    "execution_count": null,
 582    "id": "ad5dae6c-1269-4674-a49e-2efe8b956911",
 583    "metadata": {},
 584    "outputs": [],
 585    "source": []
 586   }
 587  ],
 588  "metadata": {
 589   "kernelspec": {
 590    "display_name": "Python 3 (ipykernel)",
 591    "language": "python",
 592    "name": "python3"
 593   },
 594   "language_info": {
 595    "codemirror_mode": {
 596     "name": "ipython",
 597     "version": 3
 598    },
 599    "file_extension": ".py",
 600    "mimetype": "text/x-python",
 601    "name": "python",
 602    "nbconvert_exporter": "python",
 603    "pygments_lexer": "ipython3",
 604    "version": "3.12.5"
 605   }
 606  },
 607  "nbformat": 4,
 608  "nbformat_minor": 5
 609 }