def evaluate_transfer():
evaluation = {}
for name in subjects:
def transfer_filter(paths):
return [path for path in paths if path.split("_-_")[1] == name]
transfer_model = tf.keras.models.load_model(
os.path.join(
cwd, "output"
"model_transfer_{}_{}_{}.h5".format(name, args.window,
args.band)))
single_eval = test_model(model=transfer_model,
eval_location="",
data_folder=data_folder,
batch_size=args.batch,
time_window=args.window,
batch_equalizer=Default2EnvBatchEqualizer(),
filters=[transfer_filter],
save=False)
evaluation[name] = single_eval[name]
with open(
"output/evaluation_transfer_{}_{}.json".format(
args.window, args.band), "w") as fp:
json.dump(evaluation, fp)
def evaluate():
parser = argparse.ArgumentParser(description='Evaluate the recurrent model')
parser.add_argument('--window', '--w', metavar='TIME_WINDOW', type=int,
help='size of window to train the model on', required=True)
parser.add_argument('--batch', '--b', metavar='BATCH_SIZE', type=int,
help='size of batch to use during training', required=True)
parser.add_argument('--band', '--B', metavar='BAND', type=str, choices=['full', 'delta', 'theta', 'delta_theta'],
help='which band(s) to use for testing', required=True)
args = parser.parse_args()
cwd = os.path.dirname(os.path.abspath(__file__))
root = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
test_model(model=tf.keras.models.load_model(os.path.join(cwd, "output",
"model_{}_{}.h5".format(args.window, args.band))),
eval_location="output/evaluation_{}_{}.json".format(args.window, args.band),
data_folder=os.path.join(root, "dataset", args.band), batch_size=args.batch, time_window=args.window,
batch_equalizer=Default2EnvBatchEqualizer())
train_model(
model=model,
epochs=args.epochs,
model_location=os.path.join(
cwd, "output",
"model_general_{}_{}_{}.h5".format(name, args.window,
args.band)),
log_location=os.path.join(
cwd, "output",
"training_general_{}_{}_{}.log".format(name, args.window,
args.band)),
data_folder=data_folder,
batch_size=args.batch,
time_window=args.window,
batch_equalizer=Default2EnvBatchEqualizer(),
filters=[subject_filter])
# Transfer learning
model.get_layer(name="eeg_conv1d_space").trainable = False
model.get_layer(name="eeg_bilstm").trainable = False
model.get_layer(name="env_conv1d").trainable = False
model.get_layer(name="env_lstm").trainable = False
model.get_layer(name="sigmoid").trainable = False
def transfer_filter(paths):
return [path for path in paths if path.split("_-_")[1] == name]
train_model(model=model,
epochs=args.epochs,
model_location=os.path.join(