Wrong config option in sample code

[danielschaeufele]

In line 150 of samples/mlp_learning_an_image.cu the following code is used to set the number of layers:

{"n_layers", 4},

However, this is ignored and the default is used instead, as the correct name for this option is n_hidden_layers. Also this doesn't give any warnings as unknown options are just ignored.

PS: A proper documentation for the config would be very nice, as currently it is necessary to look for the available options in the code. Also the possibility to print the used configuration would be helpful for debugging and avoiding errors like these.

[Tom94]

Good catch, thanks! Will fix that one up.

There is a documentation, actually. It's linked in the README: https://github.com/NVlabs/tiny-cuda-nn/blob/master/DOCUMENTATION.md

As for serializing / printing out the current configuration. Agreed! To a certain degree, this is already supported. The Optimizer objects have

virtual json serialize() const;

which currently conflates writing out the config and serializing the current state of the optimizer.
As time permits, I'd be totally open for a more generic to_json interface (as per the nlohmann specs).

[danielschaeufele]

Thanks for the link, I should have read the readme more carefully 😓

[Tom94]

Addressed with 8e1f4a2 :)

(Gonna keep track of the to_json todo separately. We should totally add it eventually.)