enjoy a pre-trained model after training is done?

[erwincoumans]

Is there a way to enjoy the pre-trained model (weights/checkpoint) after the training is completed?

Usually (TF Agents, OpenAI Baselines, ES Tool etc) there is a 'train' script that saves the weights of the neural network, and a 'enjoy' script that reads those weights and allows to run the pretrained network.

If it is not available, do you have any hints how it can be added?

[pat-coady]

Erwin, It's relatively straightforward to add a provision to save, load and run a model. It is slightly more involved than usual here since there are 2 NNs (value function, policy) and also the scaler coefficients. Normally I'd volunteer to make the improvement, but I am swamped with other projects at this time. I'd really like to do an implementation of AlphaGo Zero. tensorflow has a saver object, which makes saving and loading weights fairly easy. I guess the scaler coefficients would need to be saved in a little text file. Numpy has a handy save, restore method. Pat

…

On Wed, Nov 29, 2017 at 7:07 PM, erwincoumans ***@***.***> wrote: Is there a way to enjoy the pre-trained model (weights/checkpoint) after the training is completed? Usually (TF Agents, OpenAI Baselines, ES Tool etc) there is a 'train' script that saves the weights of the neural network, and a 'enjoy' script that reads those weights and allows to run the pretrained network. If it is not available, do you have any hints how it can be added? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#12>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AWdFxO7SBbhloRhKuw5d9zQd2D6aoq1iks5s7g3igaJpZM4Qv6sU> .

[llecam]

Hi,
@erwincoumans I would be interested in writing an "enjoy" script as well. Did you manage to do it since you posted this message ?

Thanks!

Léa

[erwincoumans]

I haven't looked into it further. The TensorFlow inspect_checkpoint utility can extract weights from a checkpoint:
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/tools/inspect_checkpoint.py
Please report back if you make progress.

[pat-coady]

I've added a note to the README that videos are saved to /tmp as training progresses.

I agree it would be nice to save checkpoints so that training can continue or agents can be viewed. Unfortunately, any free time I find will probably be spent on a new project.