"""
TensorFlow Model Wrapper
--------------------------
"""
import numpy as np
from .model_wrapper import ModelWrapper
[docs]class TensorFlowModelWrapper(ModelWrapper):
"""Loads a TensorFlow model and tokenizer.
TensorFlow models can use many different architectures and
tokenization strategies. This assumes that the model takes an
np.array of strings as input and returns a tf.Tensor of outputs, as
is typical with Keras modules. You may need to subclass this for
models that have dedicated tokenizers or otherwise take input
differently.
"""
def __init__(self, model):
self.model = model
def __call__(self, text_input_list, **kwargs):
text_array = np.array(text_input_list)
preds = self.model(text_array)
return preds.numpy()
[docs] def get_grad(self, text_input):
raise NotImplementedError()