-
Notifications
You must be signed in to change notification settings - Fork 0
/
autoformer_baseline.py
172 lines (145 loc) · 5.43 KB
/
autoformer_baseline.py
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
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
"""
Train similarly configured Autoformer to act as baseline for the modifications
"""
import os
import datetime
import numpy as np
import optuna
import cortical_utils
from cortical_utils import Modality
import scipy.io as sio
import argparse
parser = argparse.ArgumentParser(
prog="Cortical Response - Autoformer baseline"
)
parser.add_argument("-d", "--devices",
nargs="+", default=[0],
type=int, help="GPU to be used")
parser.add_argument("-s", "--study",
action="store", required=True,
type=str, help="Name of the study")
args = parser.parse_args()
os.environ["CUDA_VISIBLE_DEVICES"] = ",".join(
[str(dev) for dev in args.devices])
os.environ["KERAS_BACKEND"] = "jax"
import keras # pylint: disable=wrong-import-position
# pylint: disable=wrong-import-position
from keras_transformer import generic_transformer
DATA_ROOT = "./data/"
DATABASE = "sqlite:///" + DATA_ROOT + "studies.db"
STUDY_NAME = args.study
W_LEN = 9
RELEVANCE = 0.25
DS_FACTOR = 1
O = round(64 / DS_FACTOR)
STRIDE = round(np.log2(O))
EPOCHS = 200
PATIENCE = round(EPOCHS / 10)
VAL_SPLIT = 0.3
N_RUNS = 10
study = optuna.load_study(study_name=STUDY_NAME,
storage=DATABASE)
model_params = study.best_trial.params
model_params.update({
"O": O, "d": 2, "d_out": 1,
"output_components": False,
"output_attention": False,
"embed_type": "None",
"manual_dec_input": False
})
model_params["tau"] = model_params[f"tau_{max(model_params["N"],
model_params["M"]) - 1}"]
angle, comp = cortical_utils.matfile2array(DATA_ROOT)
angle = cortical_utils.preprocess_array(array=angle,
w_len=W_LEN,
relevance_factor=RELEVANCE,
downsample_factor=DS_FACTOR,
normalize=True)
comp = cortical_utils.preprocess_array(array=comp,
w_len=W_LEN,
relevance_factor=RELEVANCE,
downsample_factor=DS_FACTOR,
normalize=True)
model = generic_transformer.create_autoformer_model(**model_params)
model.compile(loss="mse", optimizer=keras.optimizers.Adam(1e-3))
model.summary()
if not os.path.exists(os.path.join(
DATA_ROOT,
"autoformer_baseline.weights.h5"
)):
split_idx = round((1 - VAL_SPLIT) * angle.shape[1])
x_enc_train, xm_enc_train, _, xm_dec_train, y_train = cortical_utils.array2io(
angle=angle[:, :split_idx, ...],
comp=comp[:, :split_idx, ...],
l_input=model_params["I"],
l_output=64,
stride=STRIDE,
modality=Modality.PREDICTION,
preserve_all=False)
x_enc_val, xm_enc_val, _, xm_dec_val, y_val = cortical_utils.array2io(
angle=angle[:, split_idx:, ...],
comp=comp[:, split_idx:, ...],
l_input=model_params["I"],
l_output=64,
stride=STRIDE,
modality=Modality.PREDICTION,
preserve_all=False)
model.fit(x=[x_enc_train, xm_enc_train, xm_dec_train],
y=y_train[..., -1, None],
batch_size=model_params["batch_size"],
epochs=EPOCHS,
shuffle=False,
validation_data=([x_enc_val, xm_enc_val, xm_dec_val],
y_val[..., -1, None]),
callbacks=[keras.callbacks.TerminateOnNaN(),
keras.callbacks.EarlyStopping(patience=PATIENCE,
mode="min"),
keras.callbacks.ModelCheckpoint(
os.path.join(DATA_ROOT,
"models",
"autoformer_baseline.weights.h5"),
monitor="val_loss",
save_best_only=True,
save_weights_only=True,
mode="min")],
verbose=1)
model.load_weights(
os.path.join(DATA_ROOT,
"models",
"autoformer_baseline.weights.h5"))
else:
model.load_weights(
os.path.join(DATA_ROOT,
"models",
"autoformer_baseline.weights.h5"))
model.summary()
original_shape = comp.shape
x_enc, xm_enc, _, xm_dec, y = cortical_utils.array2io(
angle=angle,
comp=comp,
l_input=model_params["I"],
l_output=O,
stride=O,
modality=Modality.PREDICTION,
preserve_all=True)
times = np.zeros(N_RUNS + 1, dtype="timedelta64[ms]")
for idx in range(N_RUNS + 1):
start = datetime.datetime.now(datetime.UTC)
y_pred = model.predict([x_enc, xm_enc, xm_dec],
batch_size=64, verbose=2)
times[idx] = np.timedelta64(datetime.datetime.now(datetime.UTC) - start, "ms")
# Skip warm-up: times=times[1:]
print(f"Average time: {
np.mean(times[1:].astype(np.float64)).astype("timedelta64[ms]")
} ± {
np.std(times[1:].astype(np.float64)).astype("timedelta64[ms]")
}")
y_pred = y_pred[..., -1].reshape((*original_shape[:3], -1), order="C")
y_true = y[..., -1].reshape((*original_shape[:3], -1), order="C")
x = y[..., 0].reshape((*original_shape[:3], -1), order="C")
os.makedirs(os.path.join(DATA_ROOT, "results"),
exist_ok=True, mode=0o774)
sio.savemat(os.path.join(DATA_ROOT, "results",
"autoformer_results.mat"),
{"y_true": y_true, "y_pred": y_pred, "x": x,
"model_params": model_params})