Value types for tensorflow🔗
Tensorflow is a python framework for machine learning.
Boxs comes with some value types for storing and loading tensorflow data types. These
types require, that the tensorflow package is in the PYTHONPATH and can be loaded
in order to store or load values.
TensorflowKerasModelValueType🔗
The TensorflowKerasModelValueType
allows to store a tensorflow.keras.Model. It uses the functions save_model() and
load_model() from the tensorflow.keras.models package to first save the model to
a temporary directory and then store the directory as a Zip archive. Loading the value
goes the other way round, extracting the zip to a temporary directory and then recreate
the model using load_model() function.
Supported python types🔗
None, the value type has to be used explicitly by providing it as value_type argument
to the call of store().
Configuration🔗
dir_path🔗
When a model is loaded, the path to the destination directory can be set by creating
a new TensorflowKerasModelValueType with the dir_path argument. If no dir_pathis
given, which is default, the model is loaded to a temporary directory, that
is automatically deleted. If dir_path is set, the model is extracted to the given
directory and the directory is not deleted.
default_format🔗
Additionally, the model can be stored in two different formats, 'h5' and 'tf'. As a default, 'tf' is used. For more information about this, please refer to the tensorflow documentation.
Additional meta-data attributes🔗
'model_format': The model format that was used storing the model.
TensorBoardLogDirValueType🔗
The TensorBoardLogDirValueType
allows to store the log directory for visualizing the training in Tensorboard.
Tensorboard is a web frontend that allows to display training progress and metrics.
Supported python types🔗
pathlib.Path, but the value type should only be used explicitly by providing it as
value_type argument to the call of store().
Configuration🔗
When a log directory is loaded, the path to the destination directory can be set by creating
a new TensorBoardLogDirValueType with the dir_path argument.
Additional meta-data attributes🔗
'dir_content':'tensorboard-logs'