Unpredictable quantization with quantized_bits and an alpha value of None

[qdlmcfresh]

qkeras/qkeras/quantizers.py

Line 1287 in dc0bce9

If None, the scaling factor is 1 for all channels.

states that leaving alpha at the default value of None should result in alpha being set to 1, but that's not the case which leads to unexpected behavior in quantization. The result is quantized values that are out of range a lot.

Manually setting alpha to 1 in the constructor leads to the expected results after quantization