def parse_config_common(self, arguments):
# parse batch_size, momentum, learning rate and regularization
if arguments.has_key('batch_size'):
self.batch_size = int(arguments['batch_size'])
if arguments.has_key('momentum'):
self.momentum = float(arguments['momentum'])
if arguments.has_key('lrate'):
self.lrate = parse_lrate(arguments['lrate'])
if arguments.has_key('l1_reg'):
self.l1_reg = float(arguments['l1_reg'])
if arguments.has_key('l2_reg'):
self.l2_reg = float(arguments['l2_reg'])
if arguments.has_key('max_col_norm'):
self.max_col_norm = float(arguments['max_col_norm'])
# parse activation function, including maxout
if arguments.has_key('activation'):
self.activation_text = arguments['activation']
self.activation = parse_activation(arguments['activation'])
if arguments['activation'].startswith('maxout'):
self.do_maxout = True
self.pool_size = int(arguments['activation'].replace('maxout:',''))
self.activation_text = 'maxout'
# parse dropout. note that dropout can be applied to the input features only when dropout is also
# applied to hidden-layer outputs at the same time. that is, you cannot apply dropout only to the
# input features
if arguments.has_key('dropout_factor'):
self.do_dropout = True
factors = arguments['dropout_factor'].split(',')
self.dropout_factor = [float(factor) for factor in factors]
if arguments.has_key('input_dropout_factor'):
self.input_dropout_factor = float(arguments['input_dropout_factor'])
#Added by me. Will see how this works in practice
if arguments.has_key('regression'):
self.do_regression = True
if arguments.has_key('cfg_output_file'):
self.cfg_output_file = arguments['cfg_output_file']
if arguments.has_key('param_output_file'):
self.param_output_file = arguments['param_output_file']
if arguments.has_key('kaldi_output_file'):
self.kaldi_output_file = arguments['kaldi_output_file']
if arguments.has_key('model_save_step'):
self.model_save_step = int(arguments['model_save_step'])
if arguments.has_key('non_updated_layers'):
layers = arguments['non_updated_layers'].split(",")
self.non_updated_layers = [int(layer) for layer in layers]
def parse_config_common(self, arguments):
# parse batch_size, momentum, learning rate and regularization
if arguments.has_key('batch_size'):
self.batch_size = int(arguments['batch_size'])
if arguments.has_key('momentum'):
self.momentum = float(arguments['momentum'])
if arguments.has_key('lrate'):
self.lrate = parse_lrate(arguments['lrate'])
if arguments.has_key('l1_reg'):
self.l1_reg = float(arguments['l1_reg'])
if arguments.has_key('l2_reg'):
self.l2_reg = float(arguments['l2_reg'])
if arguments.has_key('max_col_norm'):
self.max_col_norm = float(arguments['max_col_norm'])
# parse activation function, including maxout
if arguments.has_key('activation'):
self.activation_text = arguments['activation']
self.activation = parse_activation(arguments['activation'])
if arguments['activation'].startswith('maxout'):
self.do_maxout = True
self.pool_size = int(arguments['activation'].replace(
'maxout:', ''))
self.activation_text = 'maxout'
# parse dropout. note that dropout can be applied to the input features only when dropout is also
# applied to hidden-layer outputs at the same time. that is, you cannot apply dropout only to the
# input features
if arguments.has_key('dropout_factor'):
self.do_dropout = True
factors = arguments['dropout_factor'].split(',')
self.dropout_factor = [float(factor) for factor in factors]
if arguments.has_key('input_dropout_factor'):
self.input_dropout_factor = float(
arguments['input_dropout_factor'])
if arguments.has_key('cfg_output_file'):
self.cfg_output_file = arguments['cfg_output_file']
if arguments.has_key('param_output_file'):
self.param_output_file = arguments['param_output_file']
if arguments.has_key('kaldi_output_file'):
self.kaldi_output_file = arguments['kaldi_output_file']
if arguments.has_key('model_save_step'):
self.model_save_step = int(arguments['model_save_step'])
if arguments.has_key('non_updated_layers'):
layers = arguments['non_updated_layers'].split(",")
self.non_updated_layers = [int(layer) for layer in layers]
def parse_config_common(self, arguments):
# parse batch_size, momentum and learning rate
if arguments.has_key('batch_size'):
self.batch_size = int(arguments['batch_size'])
if arguments.has_key('momentum'):
self.momentum = float(arguments['momentum'])
if arguments.has_key('lrate'):
self.lrate = parse_lrate(arguments['lrate'])
# parse activation function, including maxout
if arguments.has_key('activation'):
self.activation_text = arguments['activation']
self.activation = parse_activation(arguments['activation'])
if arguments['activation'].startswith('maxout'):
self.do_maxout = True
self.pool_size = int(arguments['activation'].replace('maxout:',''))
self.activation_text = 'maxout'
# parse dropout. note that dropout can be applied to the input features only when dropout is also
# applied to hidden-layer outputs at the same time. that is, you cannot apply dropout only to the
# input features
if arguments.has_key('dropout_factor'):
self.do_dropout = True
factors = arguments['dropout_factor'].split(',')
self.dropout_factor = [float(factor) for factor in factors]
if arguments.has_key('input_dropout_factor'):
self.input_dropout_factor = float(arguments['input_dropout_factor'])
if arguments.has_key('cfg_output_file'):
self.cfg_output_file = arguments['cfg_output_file']
if arguments.has_key('param_output_file'):
self.param_output_file = arguments['param_output_file']
if arguments.has_key('kaldi_output_file'):
self.kaldi_output_file = arguments['kaldi_output_file']
if arguments.has_key('model_save_step'):
self.model_save_step = int(arguments['model_save_step'])