mtsscaling/MPIResNetExperiment.py at main · lpphd/mtsscaling · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
import warnings
import numpy as np

from utils import ScaleData

warnings.filterwarnings("ignore")

import glob
from timeit import default_timer as timer

from sklearn.metrics import f1_score, accuracy_score
from sktime_dl.deeplearning.resnet._classifier import ResNetClassifier
import pickle
from mpi4py import MPI
import pandas as pd
import os

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
import tensorflow as tf

dataset_dir_prefix = "./Datasets"

scaling_methods = ['minmax', 'maxabs', 'standard', 'robust', 'quantile', 'powert', 'normalize']
dimensions = ['timesteps', 'channels', 'both', 'all']

orig_comm = MPI.COMM_WORLD
orig_rank = orig_comm.Get_rank()

scaling_method = scaling_methods[orig_rank]

try:
    progress_list = pickle.load(open(f"resnet_{scaling_method}_progress.pkl", 'rb'))
except FileNotFoundError:
    progress_list = []

for dimension in dimensions:
    global_time = 0
    for filename in sorted(glob.glob(F"{dataset_dir_prefix}/*.npz")):
        dataset = filename.split("/")[-1].split(".")[0]

        if dataset in ['InsectWingbeat', 'CharacterTrajectories', 'JapaneseVowels', 'SpokenArabicDigits']:
            continue

        data = np.load(filename)
        orig_train_x, orig_test_x = data['train_x'].astype(np.float64), data['test_x'].astype(np.float64)
        train_y, test_y = data['train_y'], data['test_y']

        if scaling_method != "quantile":
            train_x, test_x = ScaleData(orig_train_x, orig_test_x, scaling_method, dimension, 0)
            ## Resnet requires time series length as second dimension, channels as third
            train_x = np.transpose(train_x, (0, 2, 1))
            test_x = np.transpose(test_x, (0, 2, 1))

        global_start = timer()
        for seed in range(20):

            stats = []

            if scaling_method + "_" + dimension + "_" + dataset + "_" + str(seed) in progress_list:
                print(
                    f'Skipping Dataset : {dataset} - Seed {seed} - Method: {scaling_method} - Dimension: {dimension} because it has been calculated before.')
                continue

            np.random.seed(seed)
            tf.compat.v1.set_random_seed(seed)

            if scaling_method == "quantile":
                train_x, test_x = ScaleData(orig_train_x, orig_test_x, scaling_method, dimension, seed)
                train_x = np.transpose(train_x, (0, 2, 1))
                test_x = np.transpose(test_x, (0, 2, 1))

            output_directory = f"./"

            print(
                F"Rank: {orig_rank} - Dataset : {dataset} - Seed {seed} - Method: {scaling_method} - Dimension: {dimension}",
                flush=True)
            if len(tf.config.list_physical_devices('GPU')) == 0:
                print("No GPU detected for ResNet")
                exit(-1)

            with tf.device('/device:GPU:0'):
                checkpoint_filepath = f'./{orig_rank}_best_resnet.hdf5'
                model_checkpoint_callback = tf.keras.callbacks.ModelCheckpoint(
                    filepath=checkpoint_filepath,
                    save_weights_only=True,
                    monitor='loss',
                    save_best_only=True)
                es_callback = tf.keras.callbacks.EarlyStopping(monitor='loss', min_delta=1e-4, patience=150)
                resnet = ResNetClassifier(nb_epochs=1500, verbose=False,
                                          random_state=seed,
                                          callbacks=[model_checkpoint_callback, es_callback],
                                          model_name=f"{orig_rank}_resnet",
                                          model_save_directory=output_directory)

            start = timer()
            with tf.device('/device:GPU:0'):
                resnet.fit(train_x, train_y, input_checks=False)
            end = timer()

            resnet_fitting_time = end - start

            last_epoch = len(resnet.history.history['loss'])
            resnet.model.load_weights(checkpoint_filepath)

            start = timer()
            y_pred = resnet.predict_proba(test_x).argmax(axis=1)
            end = timer()
            inference_time = end - start

            wf1 = f1_score(test_y, y_pred, average='weighted')
            acc = accuracy_score(test_y, y_pred)

            stats.append(
                [dataset, seed, resnet_fitting_time, last_epoch, inference_time,
                 acc,
                 wf1])
            stats_df = pd.DataFrame.from_records(stats, columns=['Dataset', 'Seed', 'Training Time', 'LastEpoch',
                                                                 'Inference time', 'Accuracy', 'Weighted F1'])

            stats_df.to_csv(f"resnet_uea_metrics_{scaling_method}_{dimension}.csv", mode='a', header=False,
                            index=False)
            progress_list.append(scaling_method + "_" + dimension + "_" + dataset + "_" + str(seed))
            pickle.dump(progress_list, open(F"resnet_{scaling_method}_progress.pkl", 'wb'))