I’m trying to create a neural net in numpy to make predictions on the iris dataset. I’m pretty sure I have done everything right, but in training, the model’s predicted probabilities always tend towards being almost the same for every category, and exactly the same record-to-record.

I have tried playing with learning rate, complexity (adding more nodes), and scaling the input values, and nothing seems to stop the model from tending towards making the same predictions on every row.

Rather than copy and paste all the code, I’ll just link the notebook, because I’m not sure if it even is an issue with the code, or where in the code the issue would be found:
https://github.com/mocboch/Neural-Net/blob/master/Neural%20Network%20for%20Iris%20Classification.ipynb

Thanks for your help!