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 = \"fmda_nw_202401-05_f05.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     "    time_fracs = params['time_fracs']\n",
 182     ")\n",
 183     "rnn_dat.scale_data()\n",
 184     "rnn_dat.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])"
 185    ]
 186   },
 187   {
 188    "cell_type": "code",
 189    "execution_count": null,
 190    "id": "855703c4-d7a9-4579-bca7-7c737a81d0de",
 191    "metadata": {},
 192    "outputs": [],
 193    "source": [
 194     "reproducibility.set_seed(123)\n",
 195     "rnn = RNN(params)\n",
 196     "m, errs = rnn.run_model(rnn_dat, reproducibility_run=True)"
 197    ]
 198   },
 199   {
 200    "cell_type": "markdown",
 201    "id": "49e31fdd-4c14-4a81-9e2b-4c6ba94d1f83",
 202    "metadata": {},
 203    "source": [
 204     "## Separate Models by Location"
 205    ]
 206   },
 207   {
 208    "cell_type": "code",
 209    "execution_count": null,
 210    "id": "e11e7c83-183f-48ba-abd8-a6aedff66090",
 211    "metadata": {},
 212    "outputs": [],
 213    "source": [
 214     "# Set up output dictionaries\n",
 215     "outputs_kf = {}\n",
 216     "outputs_rnn = {}"
 217    ]
 218   },
 219   {
 220    "cell_type": "code",
 221    "execution_count": null,
 222    "id": "dc5b47bd-4fbc-44b8-b2dd-d118e068b450",
 223    "metadata": {
 224     "scrolled": true
 225    },
 226    "outputs": [],
 227    "source": [
 228     "\n",
 229     "for k in param_sets_keys:\n",
 230     "    params = RNNParams(params_all[k])\n",
 231     "    print(\"~\"*80)\n",
 232     "    print(\"Running with params:\")\n",
 233     "    print(params)\n",
 234     "    # Increase Val Frac so no errors, TODO fix validation\n",
 235     "    params.update({\n",
 236     "        'train_frac': .9,\n",
 237     "        'val_frac': .05,\n",
 238     "        'activation': ['relu', 'relu'],\n",
 239     "        'epochs': 10,\n",
 240     "        'dense_units': 10,\n",
 241     "        'rnn_layers': 2       \n",
 242     "    })\n",
 243     "    for case in cases:\n",
 244     "        print(\"~\"*50)\n",
 245     "        logging.info('Processing case %s',case)\n",
 246     "        print_dict_summary(train[case])\n",
 247     "        # Format data & Run Model\n",
 248     "        # rnn_dat = create_rnn_data2(train[case], params)\n",
 249     "        rnn_dat = RNNData(train[case], scaler = params['scaler'], features_list = params['features_list'])\n",
 250     "        rnn_dat.train_test_split(\n",
 251     "            time_fracs = [.9, .05, .05]\n",
 252     "        )\n",
 253     "        rnn_dat.scale_data()\n",
 254     "        rnn_dat.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])\n",
 255     "        params.update({'bmax': rnn_dat.hours})\n",
 256     "        reproducibility.set_seed()\n",
 257     "        rnn = RNN(params)\n",
 258     "        m, errs = rnn.run_model(rnn_dat)\n",
 259     "        # Add model output to case\n",
 260     "        train[case]['m']=m\n",
 261     "        # Get RMSE Prediction Error\n",
 262     "        print(f\"RMSE: {errs}\")\n",
 263     "        outputs_rnn[case] = {'case':case, 'errs': errs.copy()}\n",
 264     "        \n",
 265     "        # Run Augmented KF\n",
 266     "        print('Running Augmented KF')\n",
 267     "        train[case]['h2'] = train[case]['hours'] // 2\n",
 268     "        train[case]['scale_fm'] = 1\n",
 269     "        m, Ec = run_augmented_kf(train[case])\n",
 270     "        m = m*rnn_dat['scale_fm']\n",
 271     "        y = rnn_dat['y']*rnn_dat['scale_fm']          \n",
 272     "        train[case]['m'] = m\n",
 273     "        print(f\"KF RMSE: {rmse(m,y)}\")\n",
 274     "        outputs_kf[case] = {'case':case, 'errs': rmse(m,y)}\n",
 275     "\n",
 276     "        # Save Outputs \n",
 277     "        to_json(outputs_rnn, \"rnn_errs.json\")\n",
 278     "        to_json(outputs_kf, \"kf_errs.json\")"
 279    ]
 280   },
 281   {
 282    "cell_type": "code",
 283    "execution_count": null,
 284    "id": "15384e4d-b8ec-4700-bdc2-83b0433d11c9",
 285    "metadata": {},
 286    "outputs": [],
 287    "source": [
 288     "logging.info('fmda_rnn_serial.ipynb done')"
 289    ]
 290   },
 291   {
 292    "cell_type": "code",
 293    "execution_count": null,
 294    "id": "d0e78fb3-b501-49d6-81a9-1a13da0134a0",
 295    "metadata": {},
 296    "outputs": [],
 297    "source": [
 298     "import importlib\n",
 299     "import moisture_rnn\n",
 300     "importlib.reload(moisture_rnn)\n",
 301     "from moisture_rnn import RNN"
 302    ]
 303   },
 304   {
 305    "cell_type": "code",
 306    "execution_count": null,
 307    "id": "37053436-8dfe-4c40-8614-811817e83782",
 308    "metadata": {},
 309    "outputs": [],
 310    "source": [
 311     "for k in outputs_rnn:\n",
 312     "    print(\"~\"*50)\n",
 313     "    print(outputs_rnn[k]['case'])\n",
 314     "    print(outputs_rnn[k]['errs']['prediction'])"
 315    ]
 316   },
 317   {
 318    "cell_type": "code",
 319    "execution_count": null,
 320    "id": "9154d5f7-015f-4ef7-af45-020410a1ea65",
 321    "metadata": {},
 322    "outputs": [],
 323    "source": [
 324     "for k in outputs_kf:\n",
 325     "    print(\"~\"*50)\n",
 326     "    print(outputs_kf[k]['case'])\n",
 327     "    print(outputs_kf[k]['errs'])"
 328    ]
 329   },
 330   {
 331    "cell_type": "markdown",
 332    "id": "f3c1c299-1655-4c64-a458-c7723db6ea6d",
 333    "metadata": {},
 334    "source": [
 335     "### TODO: FIX SCALING in Scheme below\n",
 336     "\n",
 337     "Scaling is done separately in each now."
 338    ]
 339   },
 340   {
 341    "cell_type": "markdown",
 342    "id": "0c0c3470-30f5-4915-98a7-dcdf5760d482",
 343    "metadata": {},
 344    "source": [
 345     "## Training at Multiple Locations\n",
 346     "\n",
 347     "Still sequential"
 348    ]
 349   },
 350   {
 351    "cell_type": "code",
 352    "execution_count": null,
 353    "id": "dd1aca73-7279-473e-b2a3-95aa1db7b1a8",
 354    "metadata": {},
 355    "outputs": [],
 356    "source": [
 357     "params = RNNParams(params_all['rnn'])\n",
 358     "params.update({\n",
 359     "    'epochs': 1, # less epochs since it is per location\n",
 360     "    'activation': ['relu', 'relu'],\n",
 361     "    'train_frac': .9,\n",
 362     "    'val_frac': .05,    \n",
 363     "    'dense_units': 10,\n",
 364     "    'rnn_layers': 2\n",
 365     "})\n",
 366     "\n",
 367     "# rnn_dat = create_rnn_data2(train[cases[0]], params)\n",
 368     "rnn_dat = RNNData(train[cases[0]], params['scaler'], params['features_list'])\n",
 369     "rnn_dat.train_test_split(\n",
 370     "    time_fracs = [.9, .05, .05]\n",
 371     ")\n",
 372     "rnn_dat.scale_data()\n",
 373     "rnn_dat.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])"
 374    ]
 375   },
 376   {
 377    "cell_type": "code",
 378    "execution_count": null,
 379    "id": "65b2f9a3-a8f2-4ac1-8e4d-ba38a86eaf40",
 380    "metadata": {},
 381    "outputs": [],
 382    "source": [
 383     "reproducibility.set_seed()\n",
 384     "rnn = RNN(params)"
 385    ]
 386   },
 387   {
 388    "cell_type": "code",
 389    "execution_count": null,
 390    "id": "47a85ef2-8145-4de8-9f2e-86622306ffd8",
 391    "metadata": {
 392     "scrolled": true
 393    },
 394    "outputs": [],
 395    "source": [
 396     "print(\"~\"*80)\n",
 397     "print(\"Running with params:\")\n",
 398     "print(params)\n",
 399     "\n",
 400     "for case in cases[0:10]:\n",
 401     "    print(\"~\"*50)\n",
 402     "    logging.info('Processing case %s',case)\n",
 403     "    print_dict_summary(train[case])\n",
 404     "    rnn_dat_temp = RNNData(train[case], params['scaler'], params['features_list'])\n",
 405     "    rnn_dat_temp.train_test_split(\n",
 406     "        time_fracs = [.9, .05, .05]\n",
 407     "    )\n",
 408     "    rnn_dat_temp.scale_data()\n",
 409     "    rnn_dat_temp.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])\n",
 410     "    rnn.fit(rnn_dat_temp['X_train'], rnn_dat_temp['y_train'],\n",
 411     "           validation_data=(rnn_dat_temp['X_val'], rnn_dat_temp['y_val']))\n",
 412     "    # run_rnn_pkl(train[case],param_sets[i])"
 413    ]
 414   },
 415   {
 416    "cell_type": "markdown",
 417    "id": "a0421b8d-49aa-4409-8cbf-7732f1137838",
 418    "metadata": {},
 419    "source": [
 420     "### Predict "
 421    ]
 422   },
 423   {
 424    "cell_type": "code",
 425    "execution_count": null,
 426    "id": "63d7854a-94f7-425c-9561-4fe518e044bb",
 427    "metadata": {
 428     "scrolled": true
 429    },
 430    "outputs": [],
 431    "source": [
 432     "# Predict Cases Used in Training\n",
 433     "rmses = []\n",
 434     "inds = np.arange(0,10)\n",
 435     "train_keys = list(train.keys())\n",
 436     "for i in inds:\n",
 437     "    print(\"~\"*50)\n",
 438     "    case = train_keys[i]\n",
 439     "    print(f\"Predicting case {case}\")\n",
 440     "    # rnn_dat = create_rnn_data2(train[case], params)\n",
 441     "    rnn_dat_temp = RNNData(train[case], params['scaler'], params['features_list'])\n",
 442     "    rnn_dat_temp.train_test_split(\n",
 443     "        time_fracs = [.9, .05, .05]\n",
 444     "    )\n",
 445     "    rnn_dat_temp.scale_data()\n",
 446     "    rnn_dat_temp.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])\n",
 447     "    X_temp = rnn_dat_temp.scale_all_X()\n",
 448     "    m = rnn.predict(X_temp)\n",
 449     "    test_ind = rnn_dat['test_ind']\n",
 450     "    rmses.append(rmse(m[test_ind:], rnn_dat['y_test'].flatten()))"
 451    ]
 452   },
 453   {
 454    "cell_type": "code",
 455    "execution_count": null,
 456    "id": "2a5423e0-778b-4f69-9ed0-f0082a1fefe5",
 457    "metadata": {},
 458    "outputs": [],
 459    "source": [
 460     "rmses"
 461    ]
 462   },
 463   {
 464    "cell_type": "code",
 465    "execution_count": null,
 466    "id": "45c9caae-7ced-4f21-aa05-c9b125e8fdcb",
 467    "metadata": {},
 468    "outputs": [],
 469    "source": [
 470     "pd.DataFrame({'Case': list(train.keys())[0:10], 'RMSE': rmses}).style.hide(axis=\"index\")"
 471    ]
 472   },
 473   {
 474    "cell_type": "code",
 475    "execution_count": null,
 476    "id": "f710f482-b600-4ea5-9a8a-823a13b4ec7a",
 477    "metadata": {
 478     "scrolled": true
 479    },
 480    "outputs": [],
 481    "source": [
 482     "# Predict New Locations\n",
 483     "rmses = []\n",
 484     "for i, case in enumerate(list(train.keys())[10:100]):\n",
 485     "    print(\"~\"*50)\n",
 486     "    print(f\"Predicting case {case}\")\n",
 487     "    rnn_dat_temp = RNNData(train[case], params['scaler'], params['features_list'])\n",
 488     "    rnn_dat_temp.train_test_split(\n",
 489     "        time_fracs = [.9, .05, .05]\n",
 490     "    )\n",
 491     "    rnn_dat_temp.scale_data()\n",
 492     "    rnn_dat_temp.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])\n",
 493     "    X = rnn_dat_temp.scale_all_X()\n",
 494     "    m = rnn.predict(X)\n",
 495     "    train[case]['m'] = m\n",
 496     "    test_ind = rnn_dat['test_ind']\n",
 497     "    rmses.append(rmse(m[test_ind:], rnn_dat.y_test.flatten()))\n",
 498     "\n",
 499     "df = pd.DataFrame({'Case': list(train.keys())[10:100], 'RMSE': rmses})"
 500    ]
 501   },
 502   {
 503    "cell_type": "code",
 504    "execution_count": null,
 505    "id": "d793ac87-d94b-4b16-a271-46cdc259b4fe",
 506    "metadata": {},
 507    "outputs": [],
 508    "source": [
 509     "df[0:5].style.hide(axis=\"index\")"
 510    ]
 511   },
 512   {
 513    "cell_type": "code",
 514    "execution_count": null,
 515    "id": "b99606d1-bd46-4041-8303-1bcbb196f6f4",
 516    "metadata": {},
 517    "outputs": [],
 518    "source": [
 519     "df"
 520    ]
 521   },
 522   {
 523    "cell_type": "code",
 524    "execution_count": null,
 525    "id": "52ec264d-d4b7-444c-b623-002d6383da30",
 526    "metadata": {},
 527    "outputs": [],
 528    "source": [
 529     "df.RMSE.mean()"
 530    ]
 531   },
 532   {
 533    "cell_type": "code",
 534    "execution_count": null,
 535    "id": "998922cd-46bb-4063-8284-0497e19c39b0",
 536    "metadata": {},
 537    "outputs": [],
 538    "source": [
 539     "plt.hist(df.RMSE)"
 540    ]
 541   },
 542   {
 543    "cell_type": "code",
 544    "execution_count": null,
 545    "id": "889f3bbb-9fb2-4621-9e93-1d0bc0f83e01",
 546    "metadata": {},
 547    "outputs": [],
 548    "source": []
 549   },
 550   {
 551    "cell_type": "code",
 552    "execution_count": null,
 553    "id": "fe407f61-15f2-4086-a386-7d7a5bb90d26",
 554    "metadata": {},
 555    "outputs": [],
 556    "source": []
 557   },
 558   {
 559    "cell_type": "code",
 560    "execution_count": null,
 561    "id": "2fdb63b3-68b8-4877-a7a2-f63257cb29d5",
 562    "metadata": {},
 563    "outputs": [],
 564    "source": []
 565   },
 566   {
 567    "cell_type": "code",
 568    "execution_count": null,
 569    "id": "5c7563c5-a880-45c7-8381-8ce4e1a44216",
 570    "metadata": {},
 571    "outputs": [],
 572    "source": []
 573   },
 574   {
 575    "cell_type": "code",
 576    "execution_count": null,
 577    "id": "ad5dae6c-1269-4674-a49e-2efe8b956911",
 578    "metadata": {},
 579    "outputs": [],
 580    "source": []
 581   }
 582  ],
 583  "metadata": {
 584   "kernelspec": {
 585    "display_name": "Python 3 (ipykernel)",
 586    "language": "python",
 587    "name": "python3"
 588   },
 589   "language_info": {
 590    "codemirror_mode": {
 591     "name": "ipython",
 592     "version": 3
 593    },
 594    "file_extension": ".py",
 595    "mimetype": "text/x-python",
 596    "name": "python",
 597    "nbconvert_exporter": "python",
 598    "pygments_lexer": "ipython3",
 599    "version": "3.12.5"
 600   }
 601  },
 602  "nbformat": 4,
 603  "nbformat_minor": 5
 604 }