Preparing data for inferencing

[insdrs480]

Hi there,

I have been using the load_data function to create my datasets for HuggingFace's Trainer class, which has worked really well so far. However, I am not sure how I should approach inferencing without a label column which load_data requires. Would really appreciate some advice as I am rather new to this. Thank you!

[akashsaravanan-georgian]

Hi @insdrs480, for creating datasets we expect labeled data since it follows a train-eval-test loop. If you don't have labels, you can assign dummy labels to that particular column and discard it later. Please let me know if you need any other help!

[insdrs480]

Thank you for the response! Are we able to specify the model's behaviour when it encounters unknown categorical data?

I'm getting the following error, which I believe is due to my test dataset containing values that are different from what I trained the model on: RuntimeError: mat1 and mat2 shapes cannot be multiplied (256x22 and 28x768)

[akashsaravanan-georgian]

Not at the moment unfortunately. This is something we'd like to add in though. Do you have any suggestions as to what you would like to see?

[DougTrajano]

Hi folks, I had the same issue and I suggested a solution in #61 to fix it in the load_data() function.

[akashsaravanan-georgian]

Thanks Doug!

I'm getting the following error, which I believe is due to my test dataset containing values that are different from what I trained the model on: RuntimeError: mat1 and mat2 shapes cannot be multiplied (256x22 and 28x768)

@insdrs480 Apologies I missed this bit! The model will expect the same set of columns that were used for training. You might be getting this error if they don't match. I would advice either setting up default values for missing columns or removing them from the training process entirely.

Closing this issue as Doug's fix solves this. Feel free to re-open/create a new one if you run into any issues!