def set_keras_device(device="GPU0", gpu_fraction=0.3):
    """
    Function to configure devices for Keras with the tensorflow backend
    """
     # get type of device: CPU or GPU
     device_type = device.rstrip("0123456789")
     import tensorflow as tf
     os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"
     import keras.backend as K

     if device_type == "GPU" or device_type == "gpu":
         # if no gpu is specified will use 0
         if device_type is not device:
             device_number = device[len(device_type)]
         else:
             device_number = "0"
         # set available gpu
         os.environ["CUDA_VISIBLE_DEVICES"] = device_number
         print("Setting up Keras to use GPU with miminal memory on {}".format(device_number))
         config = tf.ConfigProto()
         config.gpu_options.allow_growth = True
         config.gpu_options.per_process_gpu_memory_fraction = gpu_fraction
     elif device_type == "CPU" or device_type == "cpu":
         print("Setting up Keras to use CPU")
         config = tf.ConfigProto(device_count = {"GPU": 0})

     # set up session
     K.tensorflow_backend.set_session(tf.Session(config=config))
     sess = tf.Session(config=config)