[TB][embedding supporting] Modify histogram to accept multipy types to skip Castop and avoid OOMing in Castop
Summary: To support min/max/mean/std, SummarizeOp need to skip size checking (similar to the LpNorm error mentioned above) and accept multiple types
Test Plan:
unit test:
`buck test //caffe2/caffe2/fb/tensorboard/tests:tensorboard_accumulate_histogram_op_test`
https://our.intern.facebook.com/intern/testinfra/testrun/1407375057859572
`buck test //caffe2/caffe2/fb/tensorboard/tests:tensorboard_accumulate_histogram_op_test --stress-runs 1000`
https://our.intern.facebook.com/intern/testinfra/testrun/2533274832166362
Reviewed By: cryptopic
Differential Revision: D24605507
fbshipit-source-id: fa08372d7c9970083c38abd432d4c86e84fb10e0
diff --git a/caffe2/python/hypothesis_test.py b/caffe2/python/hypothesis_test.py
index 9298134..9e9c685 100644
--- a/caffe2/python/hypothesis_test.py
+++ b/caffe2/python/hypothesis_test.py
@@ -2715,7 +2715,7 @@
Y[X >= upper_bound] = num_buckets + 1
Y[(X >= lower_bound) & (X < upper_bound)] = \
((X[(X >= lower_bound) & (X < upper_bound)] - lower_bound) /
- segment + 1).astype(np.int32)
+ segment + 1).astype(np.int32)
for i in range(Y.shape[0]):
for j in range(Y.shape[1]):
