30 lines
878 B
Python
30 lines
878 B
Python
import numpy as np
|
|
import types
|
|
from utils import regularize
|
|
|
|
|
|
class NNLayer:
|
|
def __init__(self,
|
|
in_size: int,
|
|
out_size: int,
|
|
lr: float,
|
|
activation_func: types.FunctionType):
|
|
self.W = np.random.uniform(-1, 1, (in_size, out_size))
|
|
self.B = np.zeros((out_size))
|
|
self.lr = lr
|
|
self.last_input = None
|
|
self.last_output = None
|
|
self.activation_func = activation_func
|
|
|
|
def forward(self, V: np.ndarray) -> np.ndarray:
|
|
self.last_input = V
|
|
res = V @ self.W + self.B
|
|
self.last_output = regularize(self.activation_func(res))
|
|
return self.last_output
|
|
|
|
def backprop(self, error: np.ndarray):
|
|
dW = np.outer(self.last_input, error)
|
|
self.W -= self.lr * dW
|
|
self.B -= self.lr * error
|
|
return error @ self.W.T
|