keras_to_fnn Namespace Reference

Interface module between keras and FNN. More...

Classes
class	UnsupportedLayerException
	Exception class for unsupported layers. More...
class	UnsupportedModelException
	Exception class for unsupported models. More...

Functions
def	keras_to_txt (filename_out, model, **kwargs)
	Transforms a keras model into a txt model file which can be read by FNN. More...
def	keras_file_to_txt (filename_out, filename_in, **kwargs)
	Transforms a keras model file into a txt model file which can be read by FNN. More...

Variables
string	STR_FORMAT = '{}'.format
	Format function for strings. More...
string	INT_FORMAT = '{:>7d}'.format
	Format function for integers. More...
string	FLOAT_FORMAT = '{:0.7e}'.format
	Format function for real numbers. More...

Detailed Description

Interface module between keras and FNN.

This package provides the keras_file_to_txt and keras_to_txt functions to convert a keras model into a txt file which can be read by FNN.

Programmatic usage

Use the keras_file_to_txt function as follows to convert a .h5 file into a .txt file:

>>> keras_file_to_txt('model_out.txt', 'model_in.h5')

Use the keras_to_txt function as follows to convert a keras model into a .txt file:

>>> keras_to_txt('model_out.txt', keras_model)

Command-line usage

Use the following command to convert a .h5 file into a .txt file:

>>> python keras_to_fnn.py model_out.txt model_in.h5

Function Documentation

keras_file_to_txt()

def keras_to_fnn.keras_file_to_txt	(		filename_out,
			filename_in,
		**	kwargs
	)

Transforms a keras model file into a txt model file which can be read by FNN.

This function is actually a wrapper around keras_to_txt. It accepts the same kwargs, with the addition of custom_objects.

Note

This function uses tf.keras.models.load_model to read filename_in.

Accepted kwargs

• [in] custom_objects : list of str
  • List of custom objects to ignore while reading keras file. All other kwargs are forwarded to keras_to_txt.

    Parameters

    [in]	filename_out	The name of the txt file to write.
    [in]	filename_in	The keras model file.
    [in]	kwargs	Key-word arguments.

    Accepted kwargs
• [in] add_norm_in : bool
  • Whether to add a normalisation layer for the input.
• [in] norm_alpha_in : np.ndarray
  • Value of alpha (1d array) for the input normalisation layer, if any.
• [in] norm_beta_in : np.ndarray
  • Value of beta (1d array) for the input normalisation layer, if any.
• [in] add_norm_out : bool
  • Whether to add a normalisation layer for the output.
• [in] norm_alpha_out : np.ndarray
  • Value of alpha (1d array) for the output normalisation layer, if any.

keras_to_txt()

def keras_to_fnn.keras_to_txt	(		filename_out,
			model,
		**	kwargs
	)

Transforms a keras model into a txt model file which can be read by FNN.

Accepted kwargs

• [in] add_norm_in : bool
  • Whether to add a normalisation layer for the input.
• [in] norm_alpha_in : np.ndarray
  • Value of alpha (1d array) for the input normalisation layer, if any.
• [in] norm_beta_in : np.ndarray
  • Value of beta (1d array) for the input normalisation layer, if any.
• [in] add_norm_out : bool
  • Whether to add a normalisation layer for the output.
• [in] norm_alpha_out : np.ndarray
  • Value of alpha (1d array) for the output normalisation layer, if any.
• [in] norm_beta_out : np.ndarray
  • Value of beta (1d array) for the output normalisation layer, if any.
• [in] dropout_rates : None or float or list of float
  • Whether to add dropout after each internal layer, with the given rate.
• [in] ignore_dropout : bool
  • Whether to ignore the keras Dropout layers.

    Parameters

    [in]	filename_out	The name of the txt file to write.
    [in]	model	The keras model.
    [in]	kwargs	Key-word arguments.

Variable Documentation

FLOAT_FORMAT

string keras_to_fnn.FLOAT_FORMAT = '{:0.7e}'.format

Format function for real numbers.

INT_FORMAT

string keras_to_fnn.INT_FORMAT = '{:>7d}'.format

Format function for integers.

STR_FORMAT

string keras_to_fnn.STR_FORMAT = '{}'.format

Format function for strings.