databricks / tensorframes

[DEPRECATED] Tensorflow wrapper for DataFrames on Apache Spark
Apache License 2.0
749 stars 162 forks source link

both string and binary is not working with map_blocks #147

Open yupbank opened 6 years ago

yupbank commented 6 years ago

so string datatype is not acceptable to tensorframes' tensor and also binary can't work with map_blocks only with map_rows.

thunterdb commented 6 years ago

Yes, this is because string and byte arrays are variable length types and in tensorflow do not seem to benefit so far from a vectorized API. This could be revisited, I am sure tensorflow has improved in that respect.

Tim

On Tue, May 1, 2018 at 9:11 PM, Peng Yu notifications@github.com wrote:

so string datatype is not acceptable to tensorframes' tensor and also binary can't work with map_blocks only with map_rows.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/databricks/tensorframes/issues/147, or mute the thread https://github.com/notifications/unsubscribe-auth/AHPjAQ0tsq64-sE0Wjsxh70JY9OwZKGjks5tuMFrgaJpZM4TubtY .