Inconsistent Handling of valid Values in test_on_pod.py and pointodysseydataset_fullseq.py

[EricLina]

Issue Description:

I've noticed that in the two code files, test_on_pod.py and pointodysseydataset_fullseq.py, the valid variable is handled inconsistently:
pips2/test_on_pod.py at 8b5bd9ecb27274f76f75fcaeff0dbdf13de0b977 · aharley/pips2

    valids = d['valids'].cuda().float() # B,S,N

• In test_on_pod.py, the valid variable seems to be used directly and can contain both True and False values, which is potentially problematic since it is assumed that it should be strictly True.

pips2/datasets/pointodysseydataset_fullseq.py at 8b5bd9ecb27274f76f75fcaeff0dbdf13de0b977 · aharley/pips2

        valids = (visibs<2).astype(np.float32)

• In pointodysseydataset_fullseq.py, the valids variable is generated from the expression valids = (visibs < 2).astype(np.float32). From my review of visibs, it only contains True and False values. This means that valids should also only contain True values, leading to a mismatch in expectations between these two files.

Question:

Why is the valid variable being handled differently in these two locations? Specifically, why is the valid variable in test_on_pod.py allowed to contain both True and False values, whereas in pointodysseydataset_fullseq.py, it is assumed to only contain True values? Would it be possible to align the usage of valid across these files to ensure consistency?

Thanks in advance for your help!