38 lines
1.4 KiB
Python
Executable File
38 lines
1.4 KiB
Python
Executable File
import numpy as np
|
|
|
|
class network:
|
|
|
|
def __init__(self, inputLayerSize, *layerSizes):
|
|
if type(inputLayerSize) != int:
|
|
raise TypeError("The input layer size must be an int!")
|
|
|
|
self.weights = []
|
|
self.inputLayerSize = inputLayerSize
|
|
self.oldLayerSize = inputLayerSize
|
|
for layerSize in layerSizes:
|
|
self.weights.append( np.random.default_rng(42).random((self.oldLayerSize, layerSize)) )
|
|
self.oldLayerSize = layerSize
|
|
self.biases = [[0]*layerSize for layerSize in layerSizes]
|
|
self.weights = np.array(self.weights, dtype=object)
|
|
self.biases = np.array(self.biases, dtype=object)
|
|
|
|
def reLu(value):
|
|
return max(0, value)
|
|
|
|
def process(self, input):
|
|
if type(input) != np.ndarray:
|
|
raise TypeError("The input must be a vector!")
|
|
if input.size != self.inputLayerSize:
|
|
raise ValueError("The input vector has the wrong size!")
|
|
if input.dtype != np.float64:
|
|
raise TypeError("The input vector must contain floats!")
|
|
|
|
for layerWeights, bias in zip(self.weights, self.biases):
|
|
input = np.matmul(input, layerWeights)
|
|
input = np.add(input, bias)
|
|
#reLu application
|
|
with np.nditer(input, op_flags=['readwrite']) as layer:
|
|
for neuron in layer:
|
|
neuron = network.reLu(neuron)
|
|
|
|
return input |