Test the model for output

[joy80]

Test the model for output

[dmudiger]

Hi,

For testing on a saved model you should essentially run the same command used for training, with the following additional arguments --load-model="path to model.pt" --inference-only. Without the --inference-only argument the saved model is treated as an intermediate checkpoint and the training is resumed from this checkpoint.

For example this is the command that you'd want to use in your case:
python dlrm_s_pytorch.py --data-generation=dataset --data-set=kaggle --raw-data-file=train.txt --loss-function=bce --round-targets=True --mini-batch-size=128 --load-model=model.pt --inference-only
This should print out the test loss and accuracy.

Yes z is the output, however it would be best to print it from https://github.com/facebookresearch/dlrm/blob/master/dlrm_s_pytorch.py#L746
after the forward pass for each batch.

On a different note, the --raw-data-file is needed only for the first time when using the data-set, subsequently you can use the --processed-data-file option instead.

Hope this helps!

[joy80]

One more thing I need to know is how did you define the targets on the kaggle dataset?
In my dataset I'm getting 0 accuracy. I think is it for wrongly defined targets.

[dmudiger]

The targets are defined as 1 if clicked and 0 if not.
You can find more details here - https://www.kaggle.com/c/criteo-display-ad-challenge/data