GitLab

Add bisection based producer for increased scalability, enable fully deterministic data production, and use the materialized and bisection producer to check each other (via expected output md5's)

2nd half of rough replacement pass

fix dataset map functions

reduce bias in sample selection

cache pandas work on a daily basis

cleanup and fix batch check for multi gpu

multi device fix

fix treatment of eval data padding

print data producer

replace epoch overlap with padding and masking

move type and shape info into the producer class and update run.sh with larger batch size hyperparams

remove xla for multi GPU

more cleanup

remove model runner altogether

bug fixes

address subtle pipeline hang and improve producer __repr__

fix crash

fix assert

use popen_helper to create pools

add StreamingFilesDataset and abstract data storage to a separate class

bug fix

fix wait bug and add manual stack trace print

more bug fixes and refactor valid point mask to work with TPU sharding

misc bug fixes and adjust dtypes

address crash from decoding bools

fix remaining dtypes and change record writer pattern since it does not append

fix synthetic data

use TPUStrategy instead of TPUEstimator

minor tweaks around moving to TPUStrategy

cleanup some old code

delint and simplify permutation generation

remove low level tf layer definition, use single table with slice for keras, and misc fixes

missed minor point on removing tf layer definition

fix several bugs from recombinging layer definitions

delint and add docstrings

Update ncf_test.py. Section for identical inputs and different outputs was removed.

update data test to run against the new producer class

I've noticed sometimes the async process's pool processes do not die when ncf_main.py ends and kills the async process. This commit fixes the issue.

* intermediate commit

finish replacing spillover with resampled padding

intermediate commit

* resolve merge conflict

* intermediate commit

* further consolidate the data pipeline

* complete first pass at data pipeline refactor

* remove some leftover code

* fix test

* remove resampling, and move train padding logic into neumf.py

* small tweaks

* fix weight bug

* address PR comments

* fix dict zip. (Reed led me astray)

* delint

* make data test deterministic and delint

* Reed didn't lead me astray. I just can't read.

* more delinting

* even more delinting

* use resampling for last batch padding

* pad last batch with unique data

* Revert "pad last batch with unique data"

This reverts commit cbdf46efcd5c7907038a24105b88d38e7f1d6da2.

* move padded batch to the beginning

* delint

* fix step check for synthetic data

* Add --use_synthetic_data option to NCF.

* Add comment to _SYNTHETIC_BATCHES_PER_EPOCH

* Fix test

* Hopefully fix lint issue

* move evaluation from numpy to tensorflow

fix syntax error

don't use sigmoid to convert logits. there is too much precision loss.

WIP: add logit metrics

continue refactor of NCF evaluation

fix syntax error

fix bugs in eval loss calculation

fix eval loss reweighting

remove numpy based metric calculations

fix logging hooks

fix sigmoid to softmax bug

fix comment

catch rare PIPE error and address some PR comments

* fix metric test and address PR comments

* delint and fix python2

* fix test and address PR comments

* extend eval to TPUs

* bug fixes and add seed

* more random corrections

* make cleanup more robust

* return cleanup fn

* delint and address PR comments.

* delint and fix tests

* delinting is never done

* add pipeline hashing

* delint

* Fix spurious "did not start correctly" error.

The error "Generation subprocess did not start correctly" would occur if the async process started up after the main process checked for the subproc_alive file.

* Add error message

* Fix convergence issues for MLPerf.

Thank you to @robieta for helping me find these issues, and for providng an algorithm for the `get_hit_rate_and_ndcg_mlperf` function.

This change causes every forked process to set a new seed, so that forked processes do not generate the same set of random numbers. This improves evaluation hit rates.

Additionally, it adds a flag, --ml_perf, that makes further changes so that the evaluation hit rate can match the MLPerf reference implementation.

I ran 4 times with --ml_perf and 4 times without. Without --ml_perf, the highest hit rates achieved by each run were 0.6278, 0.6287, 0.6289, and 0.6241. With --ml_perf, the highest hit rates were 0.6353, 0.6356, 0.6367, and 0.6353.

* fix lint error

* Fix failing test

* Address @robieta's feedback

* Address more feedback

* intermediate commit

* ncf now working

* reorder pipeline

* allow batched decode for file backed dataset

* fix bug

* more tweaks

* parallize false negative generation

* shared pool hack

* workers ignore sigint

* intermediate commit

* simplify buffer backed dataset creation to fixed length record approach only. (more cleanup needed)

* more tweaks

* simplify pipeline

* fix misplaced cleanup() calls. (validation works\!)

* more tweaks

* sixify memoryview usage

* more sixification

* fix bug

* add future imports

* break up training input pipeline

* more pipeline tuning

* first pass at moving negative generation to async

* refactor async pipeline to use files instead of ipc

* refactor async pipeline

* move expansion and concatenation from reduce worker to generation workers

* abandon complete async due to interactions with the tensorflow threadpool

* cleanup

* remove per...