results_save['fn_' + pause_str] = list()
results_save['tp_' + pause_str] = list()
else:
results_save = dict()
for pause_str in pause_str_list + onset_str_list:
results_save['f_scores_' + pause_str] = list()
results_save['tn_' + pause_str] = list()
results_save['fp_' + pause_str] = list()
results_save['fn_' + pause_str] = list()
results_save['tp_' + pause_str] = list()
results_save['train_losses'], results_save['test_losses'], results_save[
'indiv_perf'], results_save['test_losses_l1'] = [], [], [], []
# %% Training
for epoch in range(0, num_epochs):
model.train()
t_epoch_strt = t.time()
loss_list = []
model.change_batch_size_reset_states(train_batch_size)
if onset_test_flag:
# setup results_dict
train_results_dict = dict()
# losses_dict = dict()
train_results_lengths = train_dataset.get_results_lengths()
for file_name in train_file_list:
# for g_f in ['g','f']:
for g_f in data_select_dict[data_set_select]:
# create new arrays for the results
train_results_dict[file_name + '/' + g_f] = np.zeros(
[train_results_lengths[file_name], prediction_length])
results_save = dict()
for pause_str in pause_str_list + overlap_str_list + onset_str_list:
results_save['f_scores_' + pause_str] = list()
results_save['tn_' + pause_str] = list()
results_save['fp_' + pause_str] = list()
results_save['fn_' + pause_str] = list()
results_save['tp_' + pause_str] = list()
results_save['train_losses'], results_save['test_losses'], results_save['indiv_perf'], results_save[
'test_losses_l1'] = [], [], [], []
# %% Training
for epoch in range(0, num_epochs):
model.train() #tells model you are in training mode, so e.g. apply dropout
t_epoch_strt = t.time()
loss_list = []
model.change_batch_size_reset_states(train_batch_size)
if onset_test_flag: #this is set to true at top of file
# setup results_dict
train_results_dict = dict()
# losses_dict = dict()
train_results_lengths = train_dataset.get_results_lengths()
for file_name in train_file_list:
for g_f in data_select_dict[data_set_select]:
# create new arrays for the onset results (the continuous predictions)
train_results_dict[file_name + '/' + g_f] = np.zeros(
[train_results_lengths[file_name], prediction_length])
train_results_dict[file_name + '/' + g_f][:] = np.nan