CODEWITHSUNDEEP
tensorflow
Yetionyo
Yetionyo

Reputation: 53

Tensorflow automatically deletes old model files

I use tf.train.Supervisor to start a session to train my model, and save model parameter every 1000 steps. But it seems that tensorflow would automatically delete old model files. Only 5 recent models were saved. Once model.ckpt-6000 is produced, the model.ckpt-1000 is deleted. But I can not find any documents on this operation.

Upvotes: 2

Views: 1681

Answers (1)

Steven
Steven

Reputation: 5162

Here are the docs: https://www.tensorflow.org/api_docs/python/tf/train/Saver

It has a variable called max_to_keep which defaults to 5. So just call 

tf.train.Saver(max_to_keep=100000)

or some other large number that encompasses all that might be created. No way to save all models as an option that I know of.

Update: as Ferran Parés points out in a comment just set max_to_keep=None to save all of the outputs.

tf.train.Saver(max_to_keep=None)

Upvotes: 3

Related Questions