From 49958d162dbc8fa9db119d2593813396f5741a6e Mon Sep 17 00:00:00 2001
From: Yannick DAYER <yannick.dayer@idiap.ch>
Date: Wed, 6 Apr 2022 10:17:13 +0200
Subject: [PATCH] Try speeding up transform by making e_step delayed
---
bob/learn/em/gmm.py | 22 ++++++++++++++++++----
1 file changed, 18 insertions(+), 4 deletions(-)
diff --git a/bob/learn/em/gmm.py b/bob/learn/em/gmm.py
index 2d5b0b9..ae7bc39 100644
--- a/bob/learn/em/gmm.py
+++ b/bob/learn/em/gmm.py
@@ -857,10 +857,24 @@ class GMMMachine(BaseEstimator):
def transform(self, X, **kwargs):
"""Returns the statistics for `X`."""
- return e_step(
- data=X,
- machine=self,
- )
+ input_is_dask, X = check_and_persist_dask_input(X)
+
+ if input_is_dask:
+ stats = [
+ dask.delayed(e_step)(
+ data=xx,
+ machine=self,
+ )
+ for xx in X
+ ]
+ stats = functools.reduce(operator.iadd, stats)
+ stats = stats.compute()
+ else:
+ stats = e_step(
+ data=X,
+ machine=self,
+ )
+ return stats
def _more_tags(self):
return {
--
GitLab