stanleyneoh / minitorch Goto Github PK

Start small

It is likely due to broadcasting changing the shape of weights and biases

Match some of the following functions

• Linear
• step
• cubic
• quadratic
• sin
• Improve test case

Makes things alot easier to work with.

(num samples, ...data)

Alternatively, create my own wrapper around matplotlib to facilitate easy OOP plotting

Such as convolutions

Specialised version of cross entropy designed just for hot vector categorical variable

perhaps a pow(a, b, abs=False) can be used.

Need to decide if __pow__(self, other) should be abs or not.

• Quite rare do we need to power a matrix by another matrix, usually some other constant. Maybe want to optimise for that.

This can be used to 'robustify' the autograd testing as well by giving custom initial