utils_mnist.py 文件源码

def get_number_of_trainable_parameters():
    """ use default graph """
    # https://stackoverflow.com/questions/38160940/ ...
    LOGGER.debug('Now compute total number of trainable params...')
    total_parameters = 0
    for variable in tf.trainable_variables():
        shape = variable.get_shape()
        name = variable.name
        variable_parameters = 1
        for dim in shape:
            variable_parameters *= dim.value
        LOGGER.debug(' layer name = {}, shape = {}, n_params = {}'.format(
            name, shape, variable_parameters
        ))
        total_parameters += variable_parameters
    LOGGER.debug('Total parameters = %d' % total_parameters)
    return total_parameters