[DOC] ValueError: For Item Prediction task a categorical_module including an item_id column is required.

Report incorrect documentation

Location of incorrect documentation

source

Describe the problems or issues found in the documentation

Code in question:

Running the code indicated above to instantiate the model results in the following traceback:

  File "/home/jason/conda/envs/candid/lib/python3.9/code.py", line 90, in runcode
    exec(code, self.locals)
  File "<input>", line 1, in <module>
  File "/home/jason/conda/envs/candid/lib/python3.9/site-packages/transformers4rec/config/transformer.py", line 58, in to_torch_model
    return torch4rec.Head(
  File "/home/jason/conda/envs/candid/lib/python3.9/site-packages/transformers4rec/torch/model/base.py", line 278, in __init__
    self.build(inputs=inputs, task_blocks=task_blocks)
  File "/home/jason/conda/envs/candid/lib/python3.9/site-packages/transformers4rec/torch/model/base.py", line 305, in build
    task.build(self.body, input_size, inputs=inputs, device=device, task_block=task_block)
  File "/home/jason/conda/envs/candid/lib/python3.9/site-packages/transformers4rec/torch/model/prediction_task.py", line 178, in build
    raise ValueError(
ValueError: For Item Prediction task a categorical_module including an item_id column is required.

I looked in the source code and it seems that is expecting there to literally be a column called item_id which is fine (but flexibility in column names would be helpful), however, I'm not sure how it fits in with the way the features were preprocessed in earlier steps and notebooks (those in examples/getting-started-session-based)where we aggregate across a session and take things like item counts and such (see below, source)

Steps taken to verify documentation is incorrect

It may not be incorrect, it is just unclear what form/shape the data should be in, in order to run NextItemPredictionTask

Suggested fix for documentation

Some kind of documentation showing what the data should look like for this task.

Report needed documentation

Report needed documentation

Describe the documentation you'd like

Steps taken to search for needed documentation**

Start with examples/getting-started-session-based/02-session-based-XLNet-with-PyT.ipynb and compare its model setup with the preprocessing in 01-ETL-with-NVTabular.ipynb. Trace how NextItemPredictionTask receives categorical features, especially the required item_id column, then document the expected data shape and update the example so it runs without the reported ValueError.