X_train,
Y_train,
batch_size=config.BATCH_SIZE,
),
#seed=config.augment_flow_seed),
epochs=config.EPOCHS,
verbose=1,
steps_per_epoch=int(1.2 * len(Y_train) / config.BATCH_SIZE),
#validation_data=datagen.flow(X_test, Y_test , batch_size=config.BATCH_SIZE),
callbacks=callbacks_list,
class_weight=config.CLASS_WEIGHT,
workers=config.WORKERS,
use_multiprocessing=config.USE_MULTIPROCESS,
max_queue_size=config.MAX_QUEUE_SIZE)
try:
best_model_path, best_accu = find_best_model(ff)
except:
best_accu = nan
trial_count += 1
seed_curr += 1
K.clear_session()
#del model
gc.collect()
expt_label = 'Run' + str(rr)
# Make soundstream of all folders
#sample_stream = make_sound_stream(file_days)
# If the selcetion table path does not exist create it
# if not os.path.exists(seltab_detect_path):
############################################################
# (1) Training classifiers using FULL dataset
# EFFECT OF RANDOM INITIALIZATION
# create a fold in cross-validation folder to store testing/evalidation results
result_path = os.path.join(config.TRAIN_RESULT_PATH, '__full_data')
#for ff in run_folders:
#for rr in range(config.NUM_RUNS):
#for rr in range(5, config.NUM_RUNS):
for rr in range(1):
ff = result_path + '/Run' + str(rr)
os.makedirs(ff, exist_ok=True)
print(ff)
best_model_path, best_accu = find_best_model(ff)
expt_label = 'Run' + str(rr)
# Make soundstream of all folders
#sample_stream = make_sound_stream(file_days)
# If the selcetion table path does not exist create it
# if not os.path.exists(seltab_detect_path):
# os.makedirs(seltab_detect_path, exist_ok=True)
print('Detecting:')
full_process_interface_dsp(expt_label, day_list, best_model_path, ff,
day_file_map, config)
K.clear_session()
gc.collect()
#cuda.select_device(0)