Search bar does not work when the metadata as missing values

Hi!

I created a dummy dataset containing 5 points as well as a file with metadata.

embeddings.tsv

-1	0	1
1	1	1
2	1	2
-1	-1	2
0	0	0

metadata.tsv

name	color	size	no
a	blue	4	21321
b	red	2	32435
c	green	6	11121
d	red	9	345435
e	red	1	12121

Loading this into TB works like a charm. However, if a column has missing values, the search bar does not work anymore at all.

To reproduce this, you can use this metadata file:

name	color	size	no
a	blue	4	123
b		2	3432432
c	green	6	453543
d	red		324354
e	red	1	1111

Can you please enhance TB, such that missing values are just ignored maybe?

Thanks a lot!

Best
Robert

Reproduce the failure in TensorBoard with the supplied embeddings.tsv and metadata.tsv examples, especially rows containing blank metadata fields. Trace the search bar's handling of metadata values and identify where blank values cause searching to stop working. Done means searches remain usable when metadata columns contain missing values, with coverage for the reproduced input.