Improvement: allow many files when using csv_input from tfx.utils.dsl_utils about tfx HOT 8 CLOSED

Thanks for the suggestion! we are working on it.

Other than supporting multiple files in the directory, we are also working on supporting new data span and data version which seems fit to your use case

from tfx.

It sounds really great, thank you!

from tfx.

I will keep this open as a feature request, Thanks!

from tfx.

Just to clarify, after 0.13.0 CSVExamplegen should be able to handle multiple CSV files for the same span and version.

@Efaq, I cannot tell from your last sentence whether you want new csv file be processed with other ones as a whole, or incrementally processing that (which falls into "continuous training" story that we have not released yet). If you think this is enough, we can improve the user guide and close this issue. Otherwise, I'd like to clarify the scope of the issue a bit further.

from tfx.

@zhitaoli The continuous training scenario is interesting, but at the moment my issue is only about having multiple csv files treated as one, so we don't need to manually join multiple csvs. So same span and version. If this is in place for 0.13.0 then it's good enough for me.

But regarding new data span and version, do you know what is in the road map?

Thank you!

from tfx.

@1025KB Can you update the docs/guide/examplegen.md to explain how to use multiple CSV files, then close this ticket? Thanks.

from tfx.

the ExampleGen tfx.utils.dsl_utils.csv_input can take folder path now,

also if you want to use regex for path, you can specify the input_config with a single split
check here for instruction

from tfx.

@zhitaoli I appreciate your work on this! However, could you explain what you mean by csv files having the same span and version? Using v0.13.0, I split the taxi data from the developer tutorial into two separate csv's, both with the same headers and schema, and attempted to load them using CsvExampleGen, but get the following RunTimeError in the x.CsvExampleGen.exec logs:

ERROR - Only one file per dir is supported

The csv_input is directed towards the appropriate folder with the data and the CsvExampleGen runs successfully with just one of the csv files present, so it appears as though it is multiple csv files that are causing the error. Do you have an idea of what might be happening?

Thanks!

from tfx.