How to save a keras model from a python recipe in a folder ?

Laurent
Laurent Registered Posts: 4 ✭✭✭✭
edited July 16 in Using Dataiku

I would like to save a keras model in a folder. I can not figure out how to save the weights of my models because I do not find the correct filepath.

The needed code to achieve this goal is :


model.save_weights(filepath)

Even with this syntax :


path = str(trained_LSTM_info['accessInfo']['root'])

model.save_weights(os.path.join(path, 'My_model_weights.h5'))

I can not use the syntax because I am on HDFS :


.get_path()

Do you have the correct syntax ?

Thank you very much.

Best Answer

  • Clément_Stenac
    Clément_Stenac Dataiker, Dataiku DSS Core Designer, Registered Posts: 753 Dataiker
    edited July 17 Answer ✓

    Hi,

    Keras can only save H5 files to a regular filesystem, not to arbitrary storage locations.

    You can either (recommended) switch your managed folder to a local folder, or:

    * Save weights to a local file

    * Then upload the models file to the managed folder using folder.upload_file(path_of_the_local_file, "path_in_the_managed_folder"). You'll need to use something like that for retrieving the file in the scoring recipe:


    with open("localfile", "wb") as out:
    with folder.get_download_stream("path-of-the-h5-file") as in:
    out.write(in.read())

Answers

Setup Info
    Tags
      Help me…