bluecamel / best_checkpoint_copier

Tensorflow Exporter that copies the best checkpoints

BestCheckpointCopier

Tensorflow Exporter that copies the best checkpoints

Usage

  best_exporter = BestCheckpointCopier(
    name='best', # directory within model directory to copy checkpoints to
    checkpoints_to_keep=10, # number of checkpoints to keep
    score_metric='Loss/total_loss', # eval_result metric to use to determine "best"
    compare_fn=lambda x,y: x.score < y.score, # comparison function used to determine "best" checkpoint (x is the current checkpoint; y is the previously copied checkpoint with the highest/worst score)
    compare_fn=lambda x,y: x.score < y.score, # comparison function used to determine "best" checkpoint
    sort_key_fn=lambda x: x.score, # key to sort on when discarding excess checkpoints
    sort_reverse=False) # sort order when discarding excess checkpoints

  final_exporter = tf.estimator.FinalExporter(
    name=final_exporter_name, serving_input_receiver_fn=predict_input_fn)

  exporters = (best_exporter, final_exporter)

  eval_specs.append(
    tf.estimator.EvalSpec(
      name=str(eval_spec_name),
      input_fn=eval_input_fn,
      steps=None,
      exporters=exporters))

best_checkpoint_copier's Issues

How to use this class?

I am interested in using your class, but I am new to the TF object detection API and I am not sure where to put your code. I am training via the default model_main_tf2.py script from the object detection repo, can I put your class there or do I have to modify the API itself?

How to initialize the function

dear, your code is a great work. But I still have a problem. Would you mind telling me how to initialize the function? Specifically, what's the x and y should be? It would be appreciated if you show detailed (practical) examples on using the function .