layers = []
relu = NN_activators.ReLUActivator()
layers.append(NN_Layer.Layer(n_inputs, n_hidden1, activator=relu))
layers.append(NN_Layer.Layer(n_hidden1, n_hidden2, activator=relu))
layers.append(NN_Layer.Layer(n_hidden2, n_outputs))
return layers
# end
housing = fetch_california_housing()
X = housing.data
y = housing.target.reshape(-1, 1)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.5,
random_state=0)
X_train = process_features(X_train)
X_test = process_features(X_test)
layers = create_layers()
loss = NN_Loss.SquaredLoss() # 回归问题，损失函数定义为平方损失函数
model = NeuralNetwork(layers, loss)
model.fit(X_train, y_train, 100000, 0.01)
y_pred = model.predict(X_test)
print("r2_score: {}.".format(r2_score(y_test, y_pred)))
n_features = 28 * 28 # 与手写图片尺寸相对应
n_hidden1 = 300 # 定义包含输出层在内的三个层
n_hidden2 = 100
n_classes = 10 # 数字分类问题，最后一层是 10个输出
layers = []
relu = NN_activators.ReLUActivator()
layers.append(NN_Layer.Layer(n_features, n_hidden1, activator=relu))
layers.append(NN_Layer.Layer(n_hidden1, n_hidden2, activator=relu))
layers.append(NN_Layer.Layer(n_hidden2, n_classes))
return layers
# end
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
X_train, Y_train = mnist.train.images, mnist.train.labels
X_test, Y_test = mnist.test.images, mnist.test.labels
layers = create_layers()
loss = NN_Loss.LogarithmicLoss()
model = NeuralNetwork(layers, loss)
model.fit(X_train, Y_train, 50000, 0.01)
V = model.predict(X_test) # 注意：输出的是与 OneHot向量同维度的一个矩阵（一个测试点对应一个 OneHot向量）
PROBA = NN_Loss.softmax(V) # 得到 softmax矩阵
y_pred = np.argmax(PROBA, axis=1)
accuracy = accuracy_score(np.argmax(Y_test, axis=1), y_pred)
print("accuracy: {}.".format(accuracy))
layers.append(NN_Layer.Layer(n_hidden1, n_hidden2, activator=relu))
layers.append(NN_Layer.Layer(n_hidden2, n_hidden3, activator=relu))
layers.append(NN_Layer.Layer(n_hidden3, n_classes, activator=ide))
return layers
# end
X, y = get_data()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=42)
encoder = MinMaxScaler()
X_train = encoder.fit_transform(X_train)
X_test = encoder.fit_transform(X_test)
encoder = OneHotEncoder()
Y_train = encoder.fit_transform(y_train) # 仅将训练数据集的标签转换成 OneHot向量构成的矩阵
layers = create_layers()
loss = NN_Loss.LogarithmicLoss() # 分类问题，使用对数损失函数（目标函数为交叉熵）
model = NeuralNetwork(layers, loss)
model.fit(X_train, Y_train, 70000, 0.02)
VR = model.predict(X_test) # 注意：输出的是与 OneHot向量同维度的一个矩阵（一个测试点对应一个 OneHot向量），可能不符合概率要求
PROBA = NN_Loss.softmax(VR) # V(R)矩阵需要经过 Softmax变换，得到 softmax矩阵，符合概率要求
y_pred = np.argmax(PROBA, axis=1) # 最大分类函数，取预测概率最大的一个类别作为最终预测结果（一个数值，0 ~ 9）
accuracy = accuracy_score(y_test, y_pred) # 计算准确率
print("accuracy: {}.".format(accuracy))