def test_mfv2_valid_indices():
for valid_indices_method in range(4):
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['model_name'] = 'multi_features_ds_v2'
input_args['output_dir'] = RUN_DIR + f'/mfv2_valid_indices_method_{valid_indices_method:02d}'
input_args['valid_indices_method'] = valid_indices_method
train_loop(input_args)
def test_mfv2_sgd():
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['model_name'] = 'multi_features_ds_v2'
input_args['output_dir'] = RUN_DIR + '/mfv2_sgd'
input_args['optimizer'] = 'sgd'
input_args['momentum'] = 0.1
train_loop(input_args)
def test_multifeatures_v2():
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['model_name'] = 'multi_features_ds_v2'
input_args['output_dir'] = RUN_DIR + '/mfv2'
input_args['signal_window_size'] = 128
input_args['mf_time_slice_interval'] = 2
input_args['subwindow_size'] = 32
train_loop(input_args)
def test_multifeatures_v2_cnn():
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['model_name'] = 'multi_features_ds_v2'
input_args['output_dir'] = RUN_DIR + '/mf_v2_cnn'
input_args['cnn_layer1_num_filters'] = 8
input_args['cnn_layer2_num_filters'] = 8
input_args['raw_num_filters'] = 0
train_loop(input_args)
def _train_loop_wrapper(rank, world_size, *train_loop_args):
torch.distributed.init_process_group('nccl',
rank=rank,
world_size=world_size)
train_loop(*train_loop_args, _rank=rank)
torch.distributed.destroy_process_group()
def test_mfv2_do_analysis():
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['model_name'] = 'multi_features_ds_v2'
input_args['output_dir'] = RUN_DIR + '/mfv2_analysis'
input_args['do_analysis'] = True
train_loop(input_args)
def test_train_loop_with_dct():
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['output_dir'] = RUN_DIR + '/with_dct'
input_args['dct_num_filters'] = 8
train_loop(input_args)
def test_train_loop_list_args():
input_args = [f"--{key}={value}" for key, value in DEFAULT_INPUT_ARGS.items()]
input_args.extend(['--output_dir', RUN_DIR + '/list_args'])
train_loop(input_args)
def test_train_loop_save_onnx():
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['output_dir'] = RUN_DIR + '/save_onnx'
input_args['save_onnx'] = True
train_loop(input_args)
def test_train_loop_raw_only():
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['output_dir'] = RUN_DIR + '/raw_only'
train_loop(input_args)
def test_multifeatures_v2_regression_with_log():
input_args = DEFAULT_INPUT_ARGS.copy()
input_args['model_name'] = 'multi_features_ds_v2'
input_args['output_dir'] = RUN_DIR + '/mf_v2_regression_with_log'
input_args['regression'] = 'log'
train_loop(input_args)
