Python/neural_network/back_propagation_neural_network.py

211 lines
5.9 KiB
Python
Raw Normal View History

2018-10-19 12:48:28 +00:00
#!/usr/bin/python
# encoding=utf8
2019-10-05 05:14:13 +00:00
"""
2018-10-19 12:48:28 +00:00
A Framework of Back Propagation Neural NetworkBP model
Easy to use:
* add many layers as you want
* clearly see how the loss decreasing
Easy to expand:
* more activation functions
* more loss functions
* more optimization method
Author: Stephen Lee
Github : https://github.com/RiptideBo
Date: 2017.11.23
2019-10-05 05:14:13 +00:00
"""
2018-10-19 12:48:28 +00:00
import numpy as np
import matplotlib.pyplot as plt
def sigmoid(x):
return 1 / (1 + np.exp(-1 * x))
2019-10-05 05:14:13 +00:00
class DenseLayer:
"""
2018-10-19 12:48:28 +00:00
Layers of BP neural network
2019-10-05 05:14:13 +00:00
"""
def __init__(
self, units, activation=None, learning_rate=None, is_input_layer=False
):
"""
2018-10-19 12:48:28 +00:00
common connected layer of bp network
:param units: numbers of neural units
:param activation: activation function
:param learning_rate: learning rate for paras
:param is_input_layer: whether it is input layer or not
2019-10-05 05:14:13 +00:00
"""
2018-10-19 12:48:28 +00:00
self.units = units
self.weight = None
self.bias = None
self.activation = activation
if learning_rate is None:
learning_rate = 0.3
self.learn_rate = learning_rate
self.is_input_layer = is_input_layer
2019-10-05 05:14:13 +00:00
def initializer(self, back_units):
self.weight = np.asmatrix(np.random.normal(0, 0.5, (self.units, back_units)))
self.bias = np.asmatrix(np.random.normal(0, 0.5, self.units)).T
2018-10-19 12:48:28 +00:00
if self.activation is None:
self.activation = sigmoid
def cal_gradient(self):
if self.activation == sigmoid:
2019-10-05 05:14:13 +00:00
gradient_mat = np.dot(self.output, (1 - self.output).T)
2018-10-19 12:48:28 +00:00
gradient_activation = np.diag(np.diag(gradient_mat))
else:
gradient_activation = 1
return gradient_activation
2019-10-05 05:14:13 +00:00
def forward_propagation(self, xdata):
2018-10-19 12:48:28 +00:00
self.xdata = xdata
if self.is_input_layer:
# input layer
self.wx_plus_b = xdata
self.output = xdata
return xdata
else:
2019-10-05 05:14:13 +00:00
self.wx_plus_b = np.dot(self.weight, self.xdata) - self.bias
2018-10-19 12:48:28 +00:00
self.output = self.activation(self.wx_plus_b)
return self.output
2019-10-05 05:14:13 +00:00
def back_propagation(self, gradient):
2018-10-19 12:48:28 +00:00
2019-10-05 05:14:13 +00:00
gradient_activation = self.cal_gradient() # i * i 维
gradient = np.asmatrix(np.dot(gradient.T, gradient_activation))
2018-10-19 12:48:28 +00:00
self._gradient_weight = np.asmatrix(self.xdata)
self._gradient_bias = -1
self._gradient_x = self.weight
2019-10-05 05:14:13 +00:00
self.gradient_weight = np.dot(gradient.T, self._gradient_weight.T)
2018-10-19 12:48:28 +00:00
self.gradient_bias = gradient * self._gradient_bias
2019-10-05 05:14:13 +00:00
self.gradient = np.dot(gradient, self._gradient_x).T
2018-10-19 12:48:28 +00:00
# ----------------------upgrade
# -----------the Negative gradient direction --------
self.weight = self.weight - self.learn_rate * self.gradient_weight
self.bias = self.bias - self.learn_rate * self.gradient_bias.T
return self.gradient
2019-10-05 05:14:13 +00:00
class BPNN:
"""
2018-10-19 12:48:28 +00:00
Back Propagation Neural Network model
2019-10-05 05:14:13 +00:00
"""
2018-10-19 12:48:28 +00:00
def __init__(self):
self.layers = []
self.train_mse = []
self.fig_loss = plt.figure()
2019-10-05 05:14:13 +00:00
self.ax_loss = self.fig_loss.add_subplot(1, 1, 1)
2018-10-19 12:48:28 +00:00
2019-10-05 05:14:13 +00:00
def add_layer(self, layer):
2018-10-19 12:48:28 +00:00
self.layers.append(layer)
def build(self):
2019-10-05 05:14:13 +00:00
for i, layer in enumerate(self.layers[:]):
2018-10-19 12:48:28 +00:00
if i < 1:
layer.is_input_layer = True
else:
2019-10-05 05:14:13 +00:00
layer.initializer(self.layers[i - 1].units)
2018-10-19 12:48:28 +00:00
def summary(self):
2019-10-05 05:14:13 +00:00
for i, layer in enumerate(self.layers[:]):
print("------- layer %d -------" % i)
print("weight.shape ", np.shape(layer.weight))
print("bias.shape ", np.shape(layer.bias))
2018-10-19 12:48:28 +00:00
2019-10-05 05:14:13 +00:00
def train(self, xdata, ydata, train_round, accuracy):
2018-10-19 12:48:28 +00:00
self.train_round = train_round
self.accuracy = accuracy
self.ax_loss.hlines(self.accuracy, 0, self.train_round * 1.1)
x_shape = np.shape(xdata)
for round_i in range(train_round):
all_loss = 0
for row in range(x_shape[0]):
2019-10-05 05:14:13 +00:00
_xdata = np.asmatrix(xdata[row, :]).T
_ydata = np.asmatrix(ydata[row, :]).T
2018-10-19 12:48:28 +00:00
# forward propagation
for layer in self.layers:
_xdata = layer.forward_propagation(_xdata)
loss, gradient = self.cal_loss(_ydata, _xdata)
all_loss = all_loss + loss
# back propagation
# the input_layer does not upgrade
for layer in self.layers[:0:-1]:
gradient = layer.back_propagation(gradient)
2019-10-05 05:14:13 +00:00
mse = all_loss / x_shape[0]
2018-10-19 12:48:28 +00:00
self.train_mse.append(mse)
self.plot_loss()
if mse < self.accuracy:
2019-10-05 05:14:13 +00:00
print("----达到精度----")
2018-10-19 12:48:28 +00:00
return mse
2019-10-05 05:14:13 +00:00
def cal_loss(self, ydata, ydata_):
self.loss = np.sum(np.power((ydata - ydata_), 2))
2018-10-19 12:48:28 +00:00
self.loss_gradient = 2 * (ydata_ - ydata)
# vector (shape is the same as _ydata.shape)
2019-10-05 05:14:13 +00:00
return self.loss, self.loss_gradient
2018-10-19 12:48:28 +00:00
def plot_loss(self):
if self.ax_loss.lines:
self.ax_loss.lines.remove(self.ax_loss.lines[0])
2019-10-05 05:14:13 +00:00
self.ax_loss.plot(self.train_mse, "r-")
2018-10-19 12:48:28 +00:00
plt.ion()
2019-10-05 05:14:13 +00:00
plt.xlabel("step")
plt.ylabel("loss")
2018-10-19 12:48:28 +00:00
plt.show()
plt.pause(0.1)
def example():
2019-10-05 05:14:13 +00:00
x = np.random.randn(10, 10)
y = np.asarray(
[
[0.8, 0.4],
[0.4, 0.3],
[0.34, 0.45],
[0.67, 0.32],
[0.88, 0.67],
[0.78, 0.77],
[0.55, 0.66],
[0.55, 0.43],
[0.54, 0.1],
[0.1, 0.5],
]
)
2018-10-19 12:48:28 +00:00
model = BPNN()
model.add_layer(DenseLayer(10))
model.add_layer(DenseLayer(20))
model.add_layer(DenseLayer(30))
model.add_layer(DenseLayer(2))
model.build()
model.summary()
2019-10-05 05:14:13 +00:00
model.train(xdata=x, ydata=y, train_round=100, accuracy=0.01)
2018-10-19 12:48:28 +00:00
2019-10-05 05:14:13 +00:00
if __name__ == "__main__":
2018-10-19 12:48:28 +00:00
example()