Commit 7f9f25a0 authored by Zhenyu Tan's avatar Zhenyu Tan Committed by TensorFlower Gardener
Browse files

Move Quantile Stream Resource to core.

Allow each Resource to manage multiple streams that share the same quantile config -- number of quantiles and epsilon. Previously each resource manage only one stream, so we will have to create resources equal to the number of features, which is cumbersome when input is high dimensional. If 1000 features use 100 quantiles (which is hardcoded today), then 1000 resources is required. This cl will create the number of resources linear to the number of parameter servers, if 2 parameter servers are present, then only 2 resources is required, one for each ps.
Remove time stamp token as the ops are called once.

PiperOrigin-RevId: 212533735
parent 4754b851
Loading
Loading
Loading
Loading
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Please to comment