abxLS submission requires double-precision floating points

[sdrobert]

Hello,

Following along with the instructions for task the Zerospeech website, it states

All numbers in the array are encoded as floats.

In the codebase, the expected data type is double-precision and will throw validation errors if that exact type is not met. While definitely floating-point, 64-bit floats are not what is usually meant by "float" (i.e. 32-bit) in PyTorch, Tensorflow, etc.

While admittedly a minor inconvenience, it would be nice if either the website specified the type or if the codebase allowed for all floating-point types.

Thank you for your time,
Sean

[nhamilakis]

I think you are right here, i think a modification of the verification function to allow all types of floats is the best solution.