机器学习（三） — 神经网络

Nueral network

1 Demand prediction

1. one feature

the activation of the price

$$a=f(x)=\frac{1}{1+e^{-(wx+b)}}$$

2. mutiple features

1. feature:
   • features: price, shipping cost, marketing, material
   • further generalization: affordability, awareness, perceived quality
2. layer: input layer, hidden layer, output layer
3. activation fuction: Sigmoid, Tanh, ReLU
4. data: transmit in the form of vector

2 Construct neural network layer

1. logic implement

param:

1. g: sigmoid function(activation fuction)
2. a: activation value

2. implement by tensorflow

x = np.array([[...]])
y = np.array([[...]])

layer_1 = Dense(units=4, activation="sigmoid")
layer_2 = Dense(units=5, activation="sigmoid")
layer_3 = Dense(units=3, activation="sigmoid")
layer_4 = Dense(units=1, activation="sigmoid")

model = Sequential([layer_1, layer_2, layer_3, layer_4])
model.compile(...)
model.fit(x, y)

model.predict(x_new)