Multilayer-Perceptron

Multilayer perception implementation by C++

How to define network structure for this mlp implemention

Using Json file to define mlp network structure

{
	"Name": "multilayer perception",
	"Data": {
		"output_num": 10,
		"type": "text",
		"file_path": "/bigData2/ycf/net.json"
	},
	"Inner": {
		"hidden_num": 5,
		"neuron_num": [
			20,
			25,
			30,
			25,
			20
		],
		"init_type": [
			"constant",
			"xavier",
			"constant",
			"xavier",
			"constant",
		],
		"type": "sigmoid"
	},
	"Loss": {
		"output_num": 10,
		"type": "softmax",
		"weights_init_type": "xavier"
	}
}

Data

using Json object to stand for input layer

• output_num: The number of input layer
• type: The form of the data, text or image
• file_path: The absolute data path

Inner

fully connected layer

• hidden_num: The number of the fully connected layer
• neuron_num: Each of hidden layers' number
• init_type: The initialization method of each hidden layers' weights
• type: Activatation function type

Loss

loss output

• output_num: The number of output layer
• type: The loss function type
• weights_init_type: The initialization method of loss layer' weights

How to difine optimization method for this mlp implemention

Using Json file to difine optimization method

{
	"net_path": "net.mlp"
	"solver": "SGD"
	"max_iter": 1000
}

• net_path:
• solver: The optimization method
• max_iter: