tfutils
tfutils copied to clipboard
neuroailab
Utilities for working with tensorflow
@nhaber * Make sure the optimizer is able to handle multiple losses * DBinterface, when things will be saved in the database * Dataset creation, graph instantiation can be done...
Slightly modify the copy deepcopy to make it support functions from class instances with tensorflow tensors as attributes.
Get Josh and Aran to contribute the loss agg_func they have implemented. @jbmelander @anayebi
For debugging as well as for book keeping purposes, it would be useful to write out the complete graph definition every time weights are saved. We can store out the...
We should have the capability of "revivifying" saved data to allow from continuing training/testing without having to pass parameters (other than the load params). This should be basically possible because...
- Allow it to be the case that saving and restoring can be to a cloud path rather than a MongoDB location
As exp_id is often not expressive enough to describe a particular experiment and changes made to an experiment, we suggest to include a mandatory experiment message that describes the most...
I've noticed that output results are not stable, meaning that different runs produce the different results, even with the same initial conditions. Specifically, even with all random seeds held constant,...