Solve the problem of TensorFlow training model and saving quantity limit
- 2021-09-12 01:36:37
- OfStack
The results of each convolution neural network training only save the last part. After consulting the relevant data under 1, it is found that it is the default value used when defining saver. Here, the following settings are made:
saver
=
tf.train.Saver(
max_to_keep
=
100
,
keep_checkpoint_every_n_hours
=
1
)
Add: Solve the problem that TensorFlow can only save 5 models
Get to the point
Find this code in the code of the training model: tf. train. Saver (),
Change to:
tf.train.Saver(max_to_keep = m) # m For the number of models you want to save
Expand
Optional parameters in the Saver class
tf.train.Saver(max_to_keep = m, keep_checkpoint_every_n_hours = n)
max_to_keep
Save the number of models closest to the current training, and the default value is 5. If you want to save it all, and the computer memory is enough, how big it is to set it.
keep_checkpoint_every_n_hours
Save the model once every n hours, with a default value of 10,000 (in general, it should not be trained for such a long time, so it is equivalent to not saving according to time, but saving according to the set number of epoch saving nodes).