| blob | 05b97f8a7f41d185225a9c83ab5ca8802ba24097 |
1 {
2 "cells": [
3 {
4 "cell_type": "markdown",
5 "id": "244c2fb0-4339-476c-a2db-a641e124e25a",
6 "metadata": {},
7 "source": [
8 "# v2.1 exploration trying to make it work better"
9 ]
10 },
11 {
12 "cell_type": "code",
13 "execution_count": null,
14 "id": "e6cc7920-e380-4b81-bac0-cd6840450e9a",
15 "metadata": {},
16 "outputs": [],
17 "source": [
18 "# Environment\n",
19 "import os\n",
20 "import os.path as osp\n",
21 "import numpy as np\n",
22 "import pandas as pd\n",
23 "import tensorflow as tf\n",
24 "import matplotlib.pyplot as plt\n",
25 "import sys\n",
26 "# Local modules\n",
27 "sys.path.append('..')\n",
28 "import reproducibility\n",
29 "import pandas as pd\n",
30 "from utils import print_dict_summary\n",
31 "from data_funcs import rmse, process_train_dict\n",
32 "from moisture_rnn import RNNParams, RNNData, RNN, RNN_LSTM\n",
33 "from moisture_rnn_pkl import pkl2train\n",
34 "from tensorflow.keras.callbacks import Callback\n",
35 "from utils import hash2\n",
36 "import copy\n",
37 "import logging\n",
38 "import pickle\n",
39 "from utils import logging_setup, read_yml, read_pkl, hash_ndarray, hash_weights\n",
40 "import yaml\n",
41 "import copy"
42 ]
43 },
44 {
45 "cell_type": "code",
46 "execution_count": null,
47 "id": "f58e8839-bf0e-4995-b966-c09e4df001ce",
48 "metadata": {},
49 "outputs": [],
50 "source": [
51 "logging_setup()"
52 ]
53 },
54 {
55 "cell_type": "markdown",
56 "id": "fae67b50-f916-45a7-bcc7-61995ba39449",
57 "metadata": {},
58 "source": [
59 "## Test Data"
60 ]
61 },
62 {
63 "cell_type": "code",
64 "execution_count": null,
65 "id": "3efed1fa-9cda-4934-8a6c-edcf179c8755",
66 "metadata": {},
67 "outputs": [],
68 "source": [
69 "file_paths = ['data/fmda_nw_202401-05_f05.pkl']"
70 ]
71 },
72 {
73 "cell_type": "code",
74 "execution_count": null,
75 "id": "28fd3746-1861-4afa-ab7e-ac449fbed322",
76 "metadata": {},
77 "outputs": [],
78 "source": [
79 "# Params used for data filtering\n",
80 "params_data = read_yml(\"params_data.yaml\") \n",
81 "params_data"
82 ]
83 },
84 {
85 "cell_type": "code",
86 "execution_count": null,
87 "id": "32a46674-8377-47f8-9c3a-e6b07f9505cf",
88 "metadata": {},
89 "outputs": [],
90 "source": [
91 "params = read_yml(\"params.yaml\", subkey='rnn') \n",
92 "params = RNNParams(params)\n",
93 "params.update({'epochs': 200, \n",
94 " 'learning_rate': 0.001,\n",
95 " 'activation': ['tanh', 'tanh'], # Activation for RNN Layers, Dense layers respectively.\n",
96 " 'recurrent_layers': 2, 'recurrent_units': 30, \n",
97 " 'dense_layers': 2, 'dense_units': 30,\n",
98 " 'early_stopping_patience': 30, # how many epochs of no validation accuracy gain to wait before stopping\n",
99 " 'batch_schedule_type': 'exp', # Hidden state batch reset schedule\n",
100 " 'bmin': 20, # Lower bound of hidden state batch reset, \n",
101 " 'bmax': params_data['hours'], # Upper bound of hidden state batch reset, using max hours\n",
102 " 'features_list': ['Ed', 'Ew', 'rain', 'elev', 'lon', 'lat', 'solar', 'wind'],\n",
103 " 'timesteps': 12\n",
104 " })"
105 ]
106 },
107 {
108 "cell_type": "code",
109 "execution_count": null,
110 "id": "91466d1b-3106-4b49-8ee8-47cbf58f938c",
111 "metadata": {
112 "scrolled": true
113 },
114 "outputs": [],
115 "source": [
116 "from data_funcs import build_train_dict2\n",
117 "train2 = build_train_dict2(file_paths, params_data, spatial=False)"
118 ]
119 },
120 {
121 "cell_type": "code",
122 "execution_count": null,
123 "id": "117dcd97-fa2a-4324-9a0e-6f7723c9c0ce",
124 "metadata": {
125 "scrolled": true
126 },
127 "outputs": [],
128 "source": [
129 "train = process_train_dict(file_paths, atm_dict=\"HRRR\", params_data = params_data, verbose=True, spatial=False)"
130 ]
131 },
132 {
133 "cell_type": "code",
134 "execution_count": null,
135 "id": "ce8e3803-44c7-4602-882c-8c69530546e7",
136 "metadata": {},
137 "outputs": [],
138 "source": [
139 "key = \"PLFI1_202401\"\n",
140 "train[key]['features_list']"
141 ]
142 },
143 {
144 "cell_type": "code",
145 "execution_count": null,
146 "id": "ca760d99-5348-4437-b30d-c5fcb252a6a1",
147 "metadata": {},
148 "outputs": [],
149 "source": [
150 "train2[key]['features_list']"
151 ]
152 },
153 {
154 "cell_type": "code",
155 "execution_count": null,
156 "id": "c3557d4b-162e-45bb-b297-4192e6b977d1",
157 "metadata": {
158 "scrolled": true
159 },
160 "outputs": [],
161 "source": [
162 "for k in train:\n",
163 " print(\"~\"*50)\n",
164 " print(k)\n",
165 " print(np.all(train[k]['X'][:,-1] == train2[k]['X'][:,-1]))"
166 ]
167 },
168 {
169 "cell_type": "code",
170 "execution_count": null,
171 "id": "d378b9fd-c92d-4c76-b039-364fb10512d9",
172 "metadata": {},
173 "outputs": [],
174 "source": [
175 "reproducibility.set_seed(123)"
176 ]
177 },
178 {
179 "cell_type": "code",
180 "execution_count": null,
181 "id": "3ec92f91-90cc-4096-a091-9ec54870ea77",
182 "metadata": {},
183 "outputs": [],
184 "source": [
185 "from itertools import islice\n",
186 "train = {k: train[k] for k in islice(train, 250)}"
187 ]
188 },
189 {
190 "cell_type": "code",
191 "execution_count": null,
192 "id": "0fb060ed-1239-44da-8c9b-923ff2004e38",
193 "metadata": {},
194 "outputs": [],
195 "source": [
196 "from data_funcs import combine_nested"
197 ]
198 },
199 {
200 "cell_type": "code",
201 "execution_count": null,
202 "id": "377ff40a-c8f3-469b-b5f5-bd46b6dc1ae1",
203 "metadata": {},
204 "outputs": [],
205 "source": [
206 "d1 = RNNData(\n",
207 " combine_nested(train), # input dictionary\n",
208 " scaler=\"standard\", # data scaling type\n",
209 " features_list = params['features_list'] # features for predicting outcome\n",
210 ")\n",
211 "\n",
212 "\n",
213 "d1.train_test_split( \n",
214 " time_fracs = [.8, .1, .1], # Percent of total time steps used for train/val/test\n",
215 " space_fracs = [.8, .1, .1] # Percent of total timeseries used for train/val/test\n",
216 ")\n",
217 "d1.scale_data()\n",
218 "\n",
219 "d1.batch_reshape(\n",
220 " timesteps = params['timesteps'], # Timesteps aka sequence length for RNN input data. \n",
221 " batch_size = params['batch_size'] # Number of samples of length timesteps for a single round of grad. descent\n",
222 ")"
223 ]
224 },
225 {
226 "cell_type": "code",
227 "execution_count": null,
228 "id": "cc7ea493-6beb-43c6-a933-032d30b8415f",
229 "metadata": {},
230 "outputs": [],
231 "source": [
232 "# Update Params specific to spatial training\n",
233 "params.update({\n",
234 " 'loc_batch_reset': d1.n_seqs # Used to reset hidden state when location changes for a given batch\n",
235 "})"
236 ]
237 },
238 {
239 "cell_type": "code",
240 "execution_count": null,
241 "id": "11c2ec92-e51d-4015-88ed-5a5c3ea4a58f",
242 "metadata": {},
243 "outputs": [],
244 "source": [
245 "reproducibility.set_seed(123)\n",
246 "rnn_sp = RNN(params)\n",
247 "m, errs = rnn_sp.run_model(d1)"
248 ]
249 },
250 {
251 "cell_type": "code",
252 "execution_count": null,
253 "id": "7b17fa96-eb8f-455f-80c3-24b384ae65e7",
254 "metadata": {},
255 "outputs": [],
256 "source": [
257 "errs.mean()"
258 ]
259 },
260 {
261 "cell_type": "code",
262 "execution_count": null,
263 "id": "5403b69a-b0c2-45d1-be52-d232fcfbe7d9",
264 "metadata": {},
265 "outputs": [],
266 "source": []
267 },
268 {
269 "cell_type": "code",
270 "execution_count": null,
271 "id": "baefb163-198e-4b8c-920a-3c520eba8579",
272 "metadata": {},
273 "outputs": [],
274 "source": [
275 "from itertools import islice\n",
276 "train2 = {k: train2[k] for k in islice(train2, 250)}"
277 ]
278 },
279 {
280 "cell_type": "code",
281 "execution_count": null,
282 "id": "54b3ea85-a4ed-4b9b-8fb5-4c2669100177",
283 "metadata": {},
284 "outputs": [],
285 "source": [
286 "d2 = RNNData(\n",
287 " combine_nested(train), # input dictionary\n",
288 " scaler=\"standard\", # data scaling type\n",
289 " features_list = params['features_list'] # features for predicting outcome\n",
290 ")\n",
291 "\n",
292 "\n",
293 "d2.train_test_split( \n",
294 " time_fracs = [.8, .1, .1], # Percent of total time steps used for train/val/test\n",
295 " space_fracs = [.8, .1, .1] # Percent of total timeseries used for train/val/test\n",
296 ")\n",
297 "d2.scale_data()\n",
298 "\n",
299 "d2.batch_reshape(\n",
300 " timesteps = params['timesteps'], # Timesteps aka sequence length for RNN input data. \n",
301 " batch_size = params['batch_size'] # Number of samples of length timesteps for a single round of grad. descent\n",
302 ")\n",
303 "# Update Params specific to spatial training\n",
304 "params.update({\n",
305 " 'loc_batch_reset': d2.n_seqs # Used to reset hidden state when location changes for a given batch\n",
306 "})"
307 ]
308 },
309 {
310 "cell_type": "code",
311 "execution_count": null,
312 "id": "aa0b0073-7b0e-4871-9c87-8d19e1c49758",
313 "metadata": {},
314 "outputs": [],
315 "source": [
316 "reproducibility.set_seed(123)\n",
317 "rnn2 = RNN(params)\n",
318 "m2, errs2 = rnn2.run_model(d2)"
319 ]
320 },
321 {
322 "cell_type": "code",
323 "execution_count": null,
324 "id": "89cdbdb5-2eee-4412-92d2-d47ef3e3549e",
325 "metadata": {},
326 "outputs": [],
327 "source": [
328 "errs2.mean()"
329 ]
330 },
331 {
332 "cell_type": "code",
333 "execution_count": null,
334 "id": "6b045231-710c-452a-bfc4-214d5e148cd8",
335 "metadata": {},
336 "outputs": [],
337 "source": []
338 },
339 {
340 "cell_type": "code",
341 "execution_count": null,
342 "id": "05184bee-a561-4541-a3e6-7dd63cb491f7",
343 "metadata": {},
344 "outputs": [],
345 "source": []
346 },
347 {
348 "cell_type": "code",
349 "execution_count": null,
350 "id": "b722325b-af18-402d-acda-daf4586e6bbc",
351 "metadata": {},
352 "outputs": [],
353 "source": []
354 },
355 {
356 "cell_type": "code",
357 "execution_count": null,
358 "id": "2685870d-6b05-4228-97f9-e017b2a4d1ee",
359 "metadata": {},
360 "outputs": [],
361 "source": []
362 },
363 {
364 "cell_type": "code",
365 "execution_count": null,
366 "id": "641aa6cd-80cf-4d62-a13c-6a7de4644778",
367 "metadata": {},
368 "outputs": [],
369 "source": []
370 },
371 {
372 "cell_type": "code",
373 "execution_count": null,
374 "id": "c110fa06-1eb4-4f24-aca4-fd852e5297c5",
375 "metadata": {},
376 "outputs": [],
377 "source": [
378 "from data_funcs import combine_nested"
379 ]
380 },
381 {
382 "cell_type": "code",
383 "execution_count": null,
384 "id": "2143ecb6-6edb-4948-8a30-698cfaceefa2",
385 "metadata": {},
386 "outputs": [],
387 "source": [
388 "nest = combine_nested(train)"
389 ]
390 },
391 {
392 "cell_type": "code",
393 "execution_count": null,
394 "id": "f8ce412b-cc9f-4273-b423-21cb53002258",
395 "metadata": {},
396 "outputs": [],
397 "source": [
398 "nest2 = combine_nested(train2)"
399 ]
400 },
401 {
402 "cell_type": "code",
403 "execution_count": null,
404 "id": "64ec6aa5-0d44-4e84-a5e7-89b2c4b4ec1a",
405 "metadata": {},
406 "outputs": [],
407 "source": [
408 "nest.keys()"
409 ]
410 },
411 {
412 "cell_type": "code",
413 "execution_count": null,
414 "id": "657e77d5-b0db-4a92-8171-0cc4e9796bb9",
415 "metadata": {},
416 "outputs": [],
417 "source": [
418 "nest2.keys()"
419 ]
420 },
421 {
422 "cell_type": "code",
423 "execution_count": null,
424 "id": "f4b2cd20-cac7-4870-8b51-2cf0726ff286",
425 "metadata": {},
426 "outputs": [],
427 "source": []
428 },
429 {
430 "cell_type": "code",
431 "execution_count": null,
432 "id": "04809b38-61af-47eb-bca5-aed80167e0ec",
433 "metadata": {},
434 "outputs": [],
435 "source": []
436 },
437 {
438 "cell_type": "code",
439 "execution_count": null,
440 "id": "8404317e-f13e-4758-82cf-07549ee9efc1",
441 "metadata": {},
442 "outputs": [],
443 "source": []
444 },
445 {
446 "cell_type": "code",
447 "execution_count": null,
448 "id": "e73de0d3-b57b-41e4-8ea1-fe7d4ac69c9b",
449 "metadata": {},
450 "outputs": [],
451 "source": []
452 },
453 {
454 "cell_type": "code",
455 "execution_count": null,
456 "id": "3582f92a-bf5b-45b7-b8ae-ea50f7ae46cd",
457 "metadata": {},
458 "outputs": [],
459 "source": []
460 },
461 {
462 "cell_type": "code",
463 "execution_count": null,
464 "id": "a54246b4-f093-4c4f-be6b-dbe9d7a8a3fd",
465 "metadata": {},
466 "outputs": [],
467 "source": []
468 },
469 {
470 "cell_type": "markdown",
471 "id": "6322f0bc-107d-40a5-96dc-804495085a99",
472 "metadata": {
473 "jp-MarkdownHeadingCollapsed": true
474 },
475 "source": [
476 "## Test Other ML"
477 ]
478 },
479 {
480 "cell_type": "code",
481 "execution_count": null,
482 "id": "12992b9a-407f-4131-ac61-e1dc338386bf",
483 "metadata": {},
484 "outputs": [],
485 "source": [
486 "params = read_yml(\"params.yaml\", subkey='xgb')\n",
487 "params"
488 ]
489 },
490 {
491 "cell_type": "code",
492 "execution_count": null,
493 "id": "f214fdf8-bb76-4912-8f8c-5d0c8c1230c2",
494 "metadata": {},
495 "outputs": [],
496 "source": [
497 "dat = read_pkl(\"data/train.pkl\")"
498 ]
499 },
500 {
501 "cell_type": "code",
502 "execution_count": null,
503 "id": "888b7805-15f6-4c09-a05b-7aed7d253f6e",
504 "metadata": {},
505 "outputs": [],
506 "source": [
507 "cases = [*dat.keys()]"
508 ]
509 },
510 {
511 "cell_type": "code",
512 "execution_count": null,
513 "id": "375055d8-c070-4639-9561-e47d3f21f1f8",
514 "metadata": {},
515 "outputs": [],
516 "source": [
517 "rnn_dat = RNNData(dat[cases[10]], params['scaler'], params['features_list'])\n",
518 "rnn_dat.train_test_split(\n",
519 " time_fracs = [.8, .1, .1]\n",
520 ")\n",
521 "rnn_dat.scale_data()"
522 ]
523 },
524 {
525 "cell_type": "code",
526 "execution_count": null,
527 "id": "e79f8dc8-5cf8-4190-b4ff-e640f61bd78b",
528 "metadata": {},
529 "outputs": [],
530 "source": [
531 "from moisture_models import XGB, RF, LM"
532 ]
533 },
534 {
535 "cell_type": "code",
536 "execution_count": null,
537 "id": "b3aeb47f-261e-4e29-9eeb-67215e5628f6",
538 "metadata": {},
539 "outputs": [],
540 "source": [
541 "mod = XGB(params)"
542 ]
543 },
544 {
545 "cell_type": "code",
546 "execution_count": null,
547 "id": "cae9a20d-1caf-45aa-a9c4-aef21b65d9c8",
548 "metadata": {},
549 "outputs": [],
550 "source": [
551 "mod.params"
552 ]
553 },
554 {
555 "cell_type": "code",
556 "execution_count": null,
557 "id": "68a07b25-c586-4fc4-a3d5-c857354e7a2c",
558 "metadata": {},
559 "outputs": [],
560 "source": [
561 "mod.fit(rnn_dat.X_train, rnn_dat.y_train)"
562 ]
563 },
564 {
565 "cell_type": "code",
566 "execution_count": null,
567 "id": "c8f88819-0a7a-4420-abb9-56a47015a4de",
568 "metadata": {},
569 "outputs": [],
570 "source": [
571 "preds = mod.predict(rnn_dat.X_test)"
572 ]
573 },
574 {
575 "cell_type": "code",
576 "execution_count": null,
577 "id": "cb7cdf14-74d6-45e4-bc1b-7d4d47dd41ac",
578 "metadata": {},
579 "outputs": [],
580 "source": [
581 "rmse(preds, rnn_dat.y_test)"
582 ]
583 },
584 {
585 "cell_type": "code",
586 "execution_count": null,
587 "id": "74d478c7-8c01-448e-9a00-dd0e1ee8e325",
588 "metadata": {},
589 "outputs": [],
590 "source": [
591 "plt.plot(rnn_dat.y_test)\n",
592 "plt.plot(preds)"
593 ]
594 },
595 {
596 "cell_type": "code",
597 "execution_count": null,
598 "id": "c5441014-c39a-4414-a779-95b81e1ed6a8",
599 "metadata": {},
600 "outputs": [],
601 "source": [
602 "params = read_yml(\"params.yaml\", subkey='rf')\n",
603 "rnn_dat = RNNData(dat[cases[10]], features_list = ['Ed', 'Ew', 'solar', 'wind', 'rain'])\n",
604 "rnn_dat.train_test_split(\n",
605 " time_fracs = [.8, .1, .1]\n",
606 ")"
607 ]
608 },
609 {
610 "cell_type": "code",
611 "execution_count": null,
612 "id": "cafe711a-20cb-4bd3-a4bc-4995a843a021",
613 "metadata": {},
614 "outputs": [],
615 "source": [
616 "import importlib\n",
617 "import moisture_models\n",
618 "importlib.reload(moisture_models)"
619 ]
620 },
621 {
622 "cell_type": "code",
623 "execution_count": null,
624 "id": "ee45f7d6-f57f-4ff6-995a-527565565f94",
625 "metadata": {},
626 "outputs": [],
627 "source": [
628 "params"
629 ]
630 },
631 {
632 "cell_type": "code",
633 "execution_count": null,
634 "id": "fafe76e5-0212-4bd1-a058-535935a08780",
635 "metadata": {},
636 "outputs": [],
637 "source": [
638 "mod2 = RF(params)\n",
639 "mod2.fit(rnn_dat.X_train, rnn_dat.y_train.flatten())\n",
640 "preds2 = mod2.predict(rnn_dat.X_test)\n",
641 "print(rmse(preds2, rnn_dat.y_test.flatten()))\n",
642 "plt.plot(rnn_dat.y_test)\n",
643 "plt.plot(preds2)"
644 ]
645 },
646 {
647 "cell_type": "code",
648 "execution_count": null,
649 "id": "c0ab4244-996c-49af-bf4a-8b0c47b0b6db",
650 "metadata": {},
651 "outputs": [],
652 "source": [
653 "from moisture_models import RF\n",
654 "mod2 = RF(params)"
655 ]
656 },
657 {
658 "cell_type": "code",
659 "execution_count": null,
660 "id": "aa6c33fd-db35-4c77-9eee-fdb39a934959",
661 "metadata": {},
662 "outputs": [],
663 "source": []
664 },
665 {
666 "cell_type": "code",
667 "execution_count": null,
668 "id": "c5598bfe-2d87-4d23-869e-aff127782462",
669 "metadata": {},
670 "outputs": [],
671 "source": [
672 "params = read_yml(\"params.yaml\", subkey='lm')\n",
673 "rnn_dat = RNNData(dat[cases[10]], features_list = ['Ed', 'Ew', 'solar', 'wind', 'rain'])\n",
674 "rnn_dat.train_test_split(\n",
675 " time_fracs = [.8, .1, .1]\n",
676 ")\n",
677 "mod = LM(params)"
678 ]
679 },
680 {
681 "cell_type": "code",
682 "execution_count": null,
683 "id": "d828c15c-4078-4967-abff-c1fd15d4696d",
684 "metadata": {},
685 "outputs": [],
686 "source": [
687 "mod.fit(rnn_dat.X_train, rnn_dat.y_train)\n",
688 "preds = mod.predict(rnn_dat.X_test)\n",
689 "print(rmse(preds2, rnn_dat.y_test.flatten()))"
690 ]
691 },
692 {
693 "cell_type": "code",
694 "execution_count": null,
695 "id": "8496a32a-8269-4d6b-953e-7f33fe626789",
696 "metadata": {},
697 "outputs": [],
698 "source": []
699 },
700 {
701 "cell_type": "code",
702 "execution_count": null,
703 "id": "75ce8bf3-6efb-4dc7-b895-def92f6ce6b4",
704 "metadata": {},
705 "outputs": [],
706 "source": []
707 },
708 {
709 "cell_type": "markdown",
710 "id": "282cb651-b21f-401d-94c5-9e07530a9ba8",
711 "metadata": {},
712 "source": [
713 "## RNN"
714 ]
715 },
716 {
717 "cell_type": "code",
718 "execution_count": null,
719 "id": "fa38f35a-d367-4df8-b2d3-7691ff4b0cf4",
720 "metadata": {},
721 "outputs": [],
722 "source": []
723 },
724 {
725 "cell_type": "markdown",
726 "id": "5ef092ff-8af1-491a-b0bf-cc3e674330e0",
727 "metadata": {},
728 "source": [
729 "## Phys Initialized"
730 ]
731 },
732 {
733 "cell_type": "code",
734 "execution_count": null,
735 "id": "5488628e-4552-4909-83e9-413fd6878bdd",
736 "metadata": {},
737 "outputs": [],
738 "source": [
739 "params.update({\n",
740 " 'epochs':100,\n",
741 " 'dense_layers': 0,\n",
742 " 'activation': ['relu', 'relu'],\n",
743 " 'phys_initialize': False,\n",
744 " 'dropout': [0,0]\n",
745 "})"
746 ]
747 },
748 {
749 "cell_type": "code",
750 "execution_count": null,
751 "id": "56bdf26c-07e7-4e4a-a567-af7dd0f564d9",
752 "metadata": {},
753 "outputs": [],
754 "source": [
755 "reproducibility.set_seed()\n",
756 "rnn = RNN(params)\n",
757 "m, errs = rnn.run_model(rnn_dat)"
758 ]
759 },
760 {
761 "cell_type": "code",
762 "execution_count": null,
763 "id": "01227b79-98f3-4931-bdfc-ff08afa8be5f",
764 "metadata": {},
765 "outputs": [],
766 "source": [
767 "rnn.model_train.summary()"
768 ]
769 },
770 {
771 "cell_type": "code",
772 "execution_count": null,
773 "id": "918a8bf0-638b-4b4b-82fe-c6a1965a72dd",
774 "metadata": {},
775 "outputs": [],
776 "source": []
777 },
778 {
779 "cell_type": "code",
780 "execution_count": null,
781 "id": "0aab34c7-8a09-480a-9d3e-619f7cf82b34",
782 "metadata": {},
783 "outputs": [],
784 "source": [
785 "params.update({\n",
786 " 'phys_initialize': True,\n",
787 " 'scaler': None, # TODO\n",
788 " 'dense_layers': 0, # NOT including single Dense output layer which is hard-coded\n",
789 " 'activation': ['linear', 'linear'], # TODO tanh, relu the same\n",
790 " 'batch_schedule_type': None # Hopefully this isn't a necessity like before, but maybe it will help\n",
791 "})"
792 ]
793 },
794 {
795 "cell_type": "code",
796 "execution_count": null,
797 "id": "ab549075-f71f-42ad-b36f-3d1e90247e33",
798 "metadata": {},
799 "outputs": [],
800 "source": [
801 "rnn_dat2 = RNNData(dat[cases[10]], params['scaler'], params['features_list'])\n",
802 "rnn_dat2.train_test_split(\n",
803 " time_fracs = [.8, .1, .1]\n",
804 ")\n",
805 "rnn_dat2.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])"
806 ]
807 },
808 {
809 "cell_type": "code",
810 "execution_count": null,
811 "id": "195f337a-ac8a-4471-8226-94863b9385e2",
812 "metadata": {},
813 "outputs": [],
814 "source": [
815 "import importlib\n",
816 "import moisture_rnn\n",
817 "importlib.reload(moisture_rnn)\n",
818 "from moisture_rnn import RNN, RNNData"
819 ]
820 },
821 {
822 "cell_type": "code",
823 "execution_count": null,
824 "id": "9395d147-17a5-44ba-aaa2-a213ffde062b",
825 "metadata": {
826 "scrolled": true
827 },
828 "outputs": [],
829 "source": [
830 "reproducibility.set_seed()\n",
831 "\n",
832 "rnn = RNN(params)"
833 ]
834 },
835 {
836 "cell_type": "code",
837 "execution_count": null,
838 "id": "d3eebe8a-ff12-454b-81b6-6a138924f127",
839 "metadata": {},
840 "outputs": [],
841 "source": [
842 "m, errs = rnn.run_model(rnn_dat2)"
843 ]
844 },
845 {
846 "cell_type": "code",
847 "execution_count": null,
848 "id": "bcbb0159-74c5-4f56-9d69-d85a58ddbd1a",
849 "metadata": {},
850 "outputs": [],
851 "source": [
852 "rnn.model_predict.get_weights()"
853 ]
854 },
855 {
856 "cell_type": "code",
857 "execution_count": null,
858 "id": "c25f741a-6280-4cf2-8017-e56672236fdb",
859 "metadata": {},
860 "outputs": [],
861 "source": []
862 },
863 {
864 "cell_type": "code",
865 "execution_count": null,
866 "id": "e8ed2b03-6123-4bdf-9e26-ef2ce4951663",
867 "metadata": {},
868 "outputs": [],
869 "source": [
870 "params['rnn_units']"
871 ]
872 },
873 {
874 "cell_type": "code",
875 "execution_count": null,
876 "id": "e44302bf-af49-4140-ae31-54f7c88a6735",
877 "metadata": {},
878 "outputs": [],
879 "source": [
880 "params.update({\n",
881 " 'phys_initialize': True,\n",
882 " 'scaler': None, # TODO\n",
883 " 'dense_layers': 0, # NOT including single Dense output layer which is hard-coded\n",
884 " 'activation': ['relu', 'relu'], # TODO tanh, relu the same\n",
885 " 'batch_schedule_type': None # Hopefully this isn't a necessity like before, but maybe it will help\n",
886 "})"
887 ]
888 },
889 {
890 "cell_type": "code",
891 "execution_count": null,
892 "id": "9a8ac32d-551c-43e8-988e-a3b13e6d9cd9",
893 "metadata": {},
894 "outputs": [],
895 "source": [
896 "rnn_dat2 = RNNData(dat[cases[10]], params['scaler'], params['features_list'])\n",
897 "rnn_dat2.train_test_split(\n",
898 " time_fracs = [.8, .1, .1]\n",
899 ")\n",
900 "rnn_dat2.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])"
901 ]
902 },
903 {
904 "cell_type": "code",
905 "execution_count": null,
906 "id": "ff727da8-38fb-4fda-999b-f712b98de0df",
907 "metadata": {
908 "scrolled": true
909 },
910 "outputs": [],
911 "source": [
912 "reproducibility.set_seed()\n",
913 "\n",
914 "rnn = RNN(params)\n",
915 "m, errs = rnn.run_model(rnn_dat2)"
916 ]
917 },
918 {
919 "cell_type": "code",
920 "execution_count": null,
921 "id": "b165074c-ea88-4b4d-8e41-6b6f22b4d221",
922 "metadata": {},
923 "outputs": [],
924 "source": []
925 },
926 {
927 "cell_type": "code",
928 "execution_count": null,
929 "id": "aa5cd4e6-4441-4c77-a086-e9edefbeb83b",
930 "metadata": {},
931 "outputs": [],
932 "source": []
933 },
934 {
935 "cell_type": "code",
936 "execution_count": null,
937 "id": "7bd1e05b-5cd8-48b4-8469-4842313d6097",
938 "metadata": {},
939 "outputs": [],
940 "source": []
941 },
942 {
943 "cell_type": "code",
944 "execution_count": null,
945 "id": "b399346d-20b8-4c97-898a-606a4be98065",
946 "metadata": {},
947 "outputs": [],
948 "source": []
949 },
950 {
951 "cell_type": "code",
952 "execution_count": null,
953 "id": "521285e6-6b6a-4d23-b688-9eb84b8eab68",
954 "metadata": {},
955 "outputs": [],
956 "source": []
957 },
958 {
959 "cell_type": "code",
960 "execution_count": null,
961 "id": "12c66af1-54fd-4398-8ee2-36eeb937c40d",
962 "metadata": {},
963 "outputs": [],
964 "source": []
965 },
966 {
967 "cell_type": "code",
968 "execution_count": null,
969 "id": "eb21fb8e-05c6-4a39-bdf1-4a57067c786d",
970 "metadata": {},
971 "outputs": [],
972 "source": []
973 },
974 {
975 "cell_type": "code",
976 "execution_count": null,
977 "id": "628a9105-ca06-44c4-ad00-13808e2f4773",
978 "metadata": {},
979 "outputs": [],
980 "source": []
981 },
982 {
983 "cell_type": "code",
984 "execution_count": null,
985 "id": "37fdbb3a-3e83-4541-93b2-982b6d4cbe93",
986 "metadata": {},
987 "outputs": [],
988 "source": []
989 },
990 {
991 "cell_type": "code",
992 "execution_count": null,
993 "id": "a592a4c9-cb3b-4174-8eaa-02afd00a1897",
994 "metadata": {},
995 "outputs": [],
996 "source": []
997 },
998 {
999 "cell_type": "code",
1000 "execution_count": null,
1001 "id": "3832fb05-417c-4648-8e2e-7748c06b3768",
1002 "metadata": {},
1003 "outputs": [],
1004 "source": []
1005 },
1006 {
1007 "cell_type": "markdown",
1008 "id": "d2360aef-e9c4-4a71-922d-336e53b82537",
1009 "metadata": {},
1010 "source": [
1011 "## LSTM\n",
1012 "\n",
1013 "TODO: FIX BELOW"
1014 ]
1015 },
1016 {
1017 "cell_type": "code",
1018 "execution_count": null,
1019 "id": "71d4e441-9bf1-4d57-bb37-091553e23212",
1020 "metadata": {},
1021 "outputs": [],
1022 "source": [
1023 "import importlib \n",
1024 "import moisture_rnn\n",
1025 "importlib.reload(moisture_rnn)\n",
1026 "from moisture_rnn import RNN_LSTM"
1027 ]
1028 },
1029 {
1030 "cell_type": "code",
1031 "execution_count": null,
1032 "id": "0f6ba896-e3be-4a9f-8a42-3df64aff7d63",
1033 "metadata": {},
1034 "outputs": [],
1035 "source": [
1036 "params = read_yml(\"params.yaml\", subkey=\"lstm\")\n",
1037 "params = RNNParams(params)"
1038 ]
1039 },
1040 {
1041 "cell_type": "code",
1042 "execution_count": null,
1043 "id": "a4cf567e-d623-4e14-b578-eed88b80d04e",
1044 "metadata": {},
1045 "outputs": [],
1046 "source": [
1047 "rnn_dat = RNNData(dat[cases[10]], params['scaler'], params['features_list'])\n",
1048 "rnn_dat.train_test_split(\n",
1049 " time_fracs = [.8, .1, .1]\n",
1050 ")\n",
1051 "rnn_dat.scale_data()\n",
1052 "rnn_dat.batch_reshape(timesteps = params['timesteps'], batch_size = params['batch_size'])"
1053 ]
1054 },
1055 {
1056 "cell_type": "code",
1057 "execution_count": null,
1058 "id": "0157a6bc-3a99-4b87-a42c-ab770d19ae37",
1059 "metadata": {},
1060 "outputs": [],
1061 "source": [
1062 "from moisture_rnn import ResetStatesCallback, EarlyStoppingCallback\n",
1063 "params.update({'epochs': 50, 'learning_rate': 0.001, 'verbose_fit': True, 'rnn_layers': 2, 'rnn_units': 20, 'dense_layers': 1, 'dense_units': 10,\n",
1064 " 'activation': ['tanh', 'tanh'], 'features_list': rnn_dat.features_list,\n",
1065 " 'batch_schedule_type':'exp', 'bmin': 10, 'bmax':rnn_dat.hours})\n",
1066 "reproducibility.set_seed(123)\n",
1067 "lstm = RNN_LSTM(params)\n",
1068 "\n",
1069 "history = lstm.model_train.fit(rnn_dat.X_train, rnn_dat.y_train, \n",
1070 " batch_size = params['batch_size'], epochs=params['epochs'], \n",
1071 " callbacks = [ResetStatesCallback(params),\n",
1072 " EarlyStoppingCallback(patience = 15)],\n",
1073 " validation_data = (rnn_dat.X_val, rnn_dat.y_val))\n",
1074 " "
1075 ]
1076 },
1077 {
1078 "cell_type": "code",
1079 "execution_count": null,
1080 "id": "ec95e7d4-6d57-441b-b673-f10625ee5dec",
1081 "metadata": {},
1082 "outputs": [],
1083 "source": []
1084 },
1085 {
1086 "cell_type": "code",
1087 "execution_count": null,
1088 "id": "9b3c8d8d-ea50-44ea-8c0c-414e07cd01ac",
1089 "metadata": {},
1090 "outputs": [],
1091 "source": []
1092 },
1093 {
1094 "cell_type": "code",
1095 "execution_count": null,
1096 "id": "03063e3c-e8f4-451d-b0cf-25bd965cd9d6",
1097 "metadata": {},
1098 "outputs": [],
1099 "source": [
1100 "params.update({'epochs': 50, 'learning_rate': 0.001, 'verbose_fit': True, 'rnn_layers': 2, 'rnn_units': 20, 'dense_layers': 1, 'dense_units': 10,\n",
1101 " 'activation': ['tanh', 'tanh'], 'features_list': rnn_dat.features_list,\n",
1102 " 'batch_schedule_type':'exp', 'bmin': 10, 'bmax':rnn_dat.hours,\n",
1103 " 'early_stopping_patience': 25})\n",
1104 "reproducibility.set_seed(123)\n",
1105 "lstm = RNN_LSTM(params)\n",
1106 "m, errs = lstm.run_model(rnn_dat)"
1107 ]
1108 },
1109 {
1110 "cell_type": "code",
1111 "execution_count": null,
1112 "id": "f60a24c6-9a67-45aa-bc5c-8818aa0ca049",
1113 "metadata": {},
1114 "outputs": [],
1115 "source": []
1116 },
1117 {
1118 "cell_type": "code",
1119 "execution_count": null,
1120 "id": "00910bd2-f050-438c-ab3b-c793b83cb5f5",
1121 "metadata": {},
1122 "outputs": [],
1123 "source": [
1124 "rnn_dat.spatial"
1125 ]
1126 },
1127 {
1128 "cell_type": "code",
1129 "execution_count": null,
1130 "id": "236b33e3-e864-4453-be16-cf07338c4105",
1131 "metadata": {},
1132 "outputs": [],
1133 "source": [
1134 "params = RNNParams(read_yml(\"params.yaml\", subkey='lstm'))\n",
1135 "params"
1136 ]
1137 },
1138 {
1139 "cell_type": "code",
1140 "execution_count": null,
1141 "id": "fe2a484c-dc99-45a9-89fc-2f451bd719b5",
1142 "metadata": {},
1143 "outputs": [],
1144 "source": [
1145 "train = read_pkl(\"data/train.pkl\")"
1146 ]
1147 },
1148 {
1149 "cell_type": "code",
1150 "execution_count": null,
1151 "id": "07bfac87-a6d4-4dcc-8d11-adf83eafab76",
1152 "metadata": {},
1153 "outputs": [],
1154 "source": [
1155 "from itertools import islice\n",
1156 "train = {k: train[k] for k in islice(train, 100)}"
1157 ]
1158 },
1159 {
1160 "cell_type": "code",
1161 "execution_count": null,
1162 "id": "4e26099b-f760-4047-afec-9e751d24b7a6",
1163 "metadata": {},
1164 "outputs": [],
1165 "source": [
1166 "from data_funcs import combine_nested\n",
1167 "rnn_dat_sp = RNNData(\n",
1168 " combine_nested(train), # input dictionary\n",
1169 " scaler=\"standard\", # data scaling type\n",
1170 " features_list = params['features_list'] # features for predicting outcome\n",
1171 ")\n",
1172 "\n",
1173 "\n",
1174 "rnn_dat_sp.train_test_split( \n",
1175 " time_fracs = [.8, .1, .1], # Percent of total time steps used for train/val/test\n",
1176 " space_fracs = [.8, .1, .1] # Percent of total timeseries used for train/val/test\n",
1177 ")\n",
1178 "rnn_dat_sp.scale_data()\n",
1179 "\n",
1180 "rnn_dat_sp.batch_reshape(\n",
1181 " timesteps = params['timesteps'], # Timesteps aka sequence length for RNN input data. \n",
1182 " batch_size = params['batch_size'] # Number of samples of length timesteps for a single round of grad. descent\n",
1183 ")"
1184 ]
1185 },
1186 {
1187 "cell_type": "code",
1188 "execution_count": null,
1189 "id": "10738795-c83b-4da3-88ba-09278caa35f8",
1190 "metadata": {},
1191 "outputs": [],
1192 "source": [
1193 "params.update({\n",
1194 " 'loc_batch_reset': rnn_dat_sp.n_seqs # Used to reset hidden state when location changes for a given batch\n",
1195 "})"
1196 ]
1197 },
1198 {
1199 "cell_type": "code",
1200 "execution_count": null,
1201 "id": "9c5d45cc-bcf0-4b6c-9c51-c4c790a2d9a5",
1202 "metadata": {},
1203 "outputs": [],
1204 "source": [
1205 "rnn_sp = RNN_LSTM(params)\n",
1206 "m_sp, errs = rnn_sp.run_model(rnn_dat_sp)"
1207 ]
1208 },
1209 {
1210 "cell_type": "code",
1211 "execution_count": null,
1212 "id": "ee332ccf-4e4a-4f66-b4d6-c079dbdb1411",
1213 "metadata": {},
1214 "outputs": [],
1215 "source": [
1216 "errs.mean()"
1217 ]
1218 },
1219 {
1220 "cell_type": "code",
1221 "execution_count": null,
1222 "id": "739d4b26-641e-47b2-a90a-67cd32215d05",
1223 "metadata": {},
1224 "outputs": [],
1225 "source": []
1226 }
1227 ],
1228 "metadata": {
1229 "kernelspec": {
1230 "display_name": "Python 3 (ipykernel)",
1231 "language": "python",
1232 "name": "python3"
1233 },
1234 "language_info": {
1235 "codemirror_mode": {
1236 "name": "ipython",
1237 "version": 3
1238 },
1239 "file_extension": ".py",
1240 "mimetype": "text/x-python",
1241 "name": "python",
1242 "nbconvert_exporter": "python",
1243 "pygments_lexer": "ipython3",
1244 "version": "3.12.5"
1245 }
1246 },
1247 "nbformat": 4,
1248 "nbformat_minor": 5
1249 }