Arnon Senderov

Following the behavior of full python within the existing types adds very little code, I'm in a process if doing it right now and I will share the fix. For...

Update, The attached script tests all the combinations of ARRAY+ARRAY and ARRAY+SCALRE, here is its partial printout. Test s/s: s8 + s8 = array([0], dtype=int8) s8 + s16 = array([0],...

I pushed code to my fork: 1. Add new types: INT32 UINT32 INT64 2. Fix upscaling to match formal Numpy. 3. Support several operators which were not supported before 4....

Thanks a lot for your comments, I agree with most of them and I will refer to each. 1. my recent fork and commit reflects work I've done for a...

please note the fix consists of two parts, the upcasting function and the array/scalar type matching.

After updating "compare" code reduction is now 26%. Regarding the FFT, mine has pre-prepared tables and it's platform specific. I fully understand your choice of no RAM overhead.