You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
If we would add the wrapper to the function, we would have to calculate the hash value for a large array, which might take a long time and might require loading the array to memory. Both of which are not desirable.
The previous approach was optimized to use the built-in hashing of dask.array (array.name). While this leads to significantly faster results, the hash is not always the same between different kernels. Currently it is not clear if that is due to the dask package, or our implementation. This will have to be revisited later.
If we would add the wrapper to the function, we would have to calculate the hash value for a large array, which might take a long time and might require loading the array to memory. Both of which are not desirable.
The text was updated successfully, but these errors were encountered: