193c2dffbf492f832d2914d16b12f7ba8afc3979,models/official/mnasnet/mnasnet_main.py,,build_model_fn,#Any#Any#Any#Any#,263

Before Change

      // expects [batch_size, ...] Tensors, thus reshape to introduce a batch
      // dimension. These Tensors are implicitly concatenated to
      // [params["batch_size"]].
      gs_t = tf.reshape(global_step, [1])
      loss_t = tf.reshape(loss, [1])
      lr_t = tf.reshape(learning_rate, [1])
      ce_t = tf.reshape(current_epoch, [1])

      host_call = (host_call_fn, [gs_t, loss_t, lr_t, ce_t])

  else:
    train_op = None

After Change

      if params["add_summaries"]:
        summary_writer = tf2.summary.create_file_writer(
            FLAGS.model_dir, max_queue=params["iterations_per_loop"])
        with summary_writer.as_default():
          should_record = tf.equal(global_step % params["iterations_per_loop"],
                                   0)
          with tf2.summary.record_if(should_record):
            tf2.summary.scalar("loss", loss, step=global_step)
            tf2.summary.scalar("learning_rate", learning_rate, step=global_step)
            tf2.summary.scalar("current_epoch", current_epoch, step=global_step)

    // Batch normalization requires UPDATE_OPS to be added as a dependency to
    // the train operation.
    update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)

    with tf.control_dependencies(update_ops + tf.summary.all_v2_summary_ops()):
      train_op = optimizer.minimize(loss, global_step)

Instances