testOpFormattedDataExport fails sometimes

[jakirkham]

The main test in testOpFormattedDataExport, testBasic, fails here sometimes. As this test takes random data (

lazyflow/tests/testOpFormattedDataExport.py

Line 47 in 090b6cb

 data = numpy.random.random( (100,100) ).astype( numpy.float32 ) * 100 

) and is asserting the validity of that data (

lazyflow/tests/testOpFormattedDataExport.py

Line 92 in 090b6cb

 assert (numpy.abs(difference_from_expected) <= 1).all(), "Read data didn't match exported data!" 

), it is possible that some types of random values are not acceptable. It would be nice to figure out the cause of this and fix it.

[jakirkham]

Actually, the cause seems to be rather straightforward. As the type we are using is unsigned (

lazyflow/tests/testOpFormattedDataExport.py

Line 82 in 090b6cb

 expected_data = expected_data.astype(numpy.uint8) 

), we are getting a rollover error (i.e. sometimes we get 255), which is not less than or equal to 1 even though the two values may have been (i.e. 1 and 2 ), but had their difference taken in a non-positive direction.

[jakirkham]

This can be easily resolved several ways. We could simply promote the type before subtraction. Alternatively, if we need to maintain the same type, we could ensure we have the larger values in one array and the smaller ones in the other.