[searchlight] sys.exit used in _get_block_data

[mihaic]

https://github.com/IntelPNI/brainiak/blob/master/brainiak/searchlight/searchlight.py#L136
Instead, an exception should be raised (and documented) in distribute if the input data does not have the proper shape.

[YSanchezAraujo]

@mihaic are you saying the exception should be raised here instead of here

[mihaic]

@YSanchezAraujo, thanks for looking into this issue.

Indeed, I am saying that an exception should be raised in distribute, such that the user code can decide what to do, instead of exiting the program as we are doing now. This is the approach we are using everywhere else in BrainIAK.

In practice, in addition to raising an exception in _get_block_data, the exception should be caught and a new one raised in distribute (so the user does not see the implementation details in the traceback).

Instead, it would be simpler to check the input shape in distribute and just remove the else code from _get_block_data.

[YSanchezAraujo]

thanks for the response, some questions:

is the mask argument guaranteed to be a numpy array, and or the 4d arrays inside subjects, should there be an instance check and maybe a np.array(<argument>) if isinstance(<argument>, np.ndarray) returns false?

thinking of something like:

assert mask.ndim == 3, "mask should be a 3D array"

for (idx, subj) in enumerate(subjects):
    assert subj.ndim == 4, \
    "expected 4D data for index {} in subjects".format(idx)

[mihaic]

We should also check mask. And arguments should be np.ndarray instances. It is better if users explicitly apply np.array to other data structures themselves and deal with potential issues.