These commits address two issues brought up in #9 related to the interactive pipeline notebook. First, the Transformer component errors out because the zip_code field in the current upstream dataset is inferred as INT and module.preprocessing_fn attempts to process it as a string. Second, the Trainer component errors out because the upstream dataset has the consumer_disputed field already converted to {0,1} but download_dataset.update_csv attempts to do that again, resulting in nans that the trainer chokes on.
I took a best guess at patches for both. I'm open for suggestions.
These commits address two issues brought up in #9 related to the interactive pipeline notebook. First, the
Transformer
component errors out because thezip_code
field in the current upstream dataset is inferred asINT
andmodule.preprocessing_fn
attempts to process it as a string. Second, theTrainer
component errors out because the upstream dataset has theconsumer_disputed
field already converted to {0,1} butdownload_dataset.update_csv
attempts to do that again, resulting innan
s that the trainer chokes on.I took a best guess at patches for both. I'm open for suggestions.