From 20eedc46418470c42ff2069c1da98df0176bda0c Mon Sep 17 00:00:00 2001
From: Yannick DAYER <yannick.dayer@idiap.ch>
Date: Thu, 10 Nov 2022 12:57:04 +0100
Subject: [PATCH] [fix] Specify the chunk size of arrays for k_init.
The "auto" chunk size somehow breaks k_init for large numpy arrays.
---
src/bob/learn/em/kmeans.py | 1 +
1 file changed, 1 insertion(+)
diff --git a/src/bob/learn/em/kmeans.py b/src/bob/learn/em/kmeans.py
index 8f968eb..bc2664f 100644
--- a/src/bob/learn/em/kmeans.py
+++ b/src/bob/learn/em/kmeans.py
@@ -308,6 +308,7 @@ class KMeansMachine(BaseEstimator):
# k_init requires da.Array as input.
logger.debug("Transform k-means data to dask array")
data = da.array(data)
+ data.rechunk(1, data.shape[-1]) # Prevents issue with large arrays.
logger.debug("Get k-means centroids")
self.centroids_ = k_init(
X=data,
--
GitLab