563207ae6bd9a8565243b04dcd92af9c33bac524,deepchem/models/keras_model.py,KerasModel,fit_generator,#KerasModel#Any#Any#Any#Any#Any#Any#Any#,246

Before Change

      if loss is None:
        loss = self._loss_fn
      with tf.GradientTape() as tape:
        if len(inputs) == 1:
          inputs = inputs[0]
        outputs = self.model(inputs)
        if isinstance(outputs, tf.Tensor):
          outputs = [outputs]
        if self._loss_outputs is not None:
          outputs = [outputs[i] for i in self._loss_outputs]
        batch_loss = loss(outputs, labels, weights)
      if variables is None:
        vars = self.model.trainable_variables
      else:
        vars = variables

After Change

      loss = self._loss_fn
    var_key = None
    if variables is not None:
      var_key = tuple(v.experimental_ref() for v in variables)

      // The optimizer creates internal variables the first time apply_gradients()
      // is called for a new set of variables.  If that happens inside a function
      // annotated with tf.function it throws an exception, so call it once here.

      zero_grads = [tf.zeros(v.shape) for v in variables]
      self._tf_optimizer.apply_gradients(zip(zero_grads, variables))
    if var_key not in self._gradient_fn_for_vars:
      self._gradient_fn_for_vars[var_key] = self._create_gradient_fn(variables)
    apply_gradient_for_batch = self._gradient_fn_for_vars[var_key]
    time1 = time.time()

    // Main training loop.

Instances