An entry artifact tensorflowplayground

tensorflow playground brief introduction

TensorFlow The amusement park is a simple neural network that can be trained through a web browser And realize the tool of visual training process .

Address ：http://playground.tensorflow.org/

TensorFlow playground Use

TensorFlow playground Interface

TensorFlow playground Interface diagram

Function details

（1） Control operation

The three functions from left to right are ：（a） restart ;（b） function ;（c） Run one cycle at a time

（2） Number of operating cycles

Used to view the number of training cycles

（3） Parameter adjustment area

name ——> Functional specifications

Learning rate ——> Learning rate （ It's a super parameter , Gradient descent algorithm will be used ; The learning rate is set artificially according to the actual situation ）.

Activation——> Activation function （ The default is nonlinear function Tanh; If for the linear classification problem , The activation function can not be used here ）.

Regularization——> Regularization （ Regularization is the use of norms to solve the problem of over fitting ）. Problem type Question type （ What we want to solve here is a binary classification problem , Simply explain that the classification problem refers to , Given a new pattern , Infer its corresponding category according to the training set （ Such as ：+1,-1）, It is a qualitative output , Also called discrete variable prediction ; The regression problem refers to , Given a new pattern , Infer its corresponding output value according to the training set （ The set of real Numbers ） How much is the , It is a quantitative output , Also called continuous variable prediction ; Here we belong to the classification problem .）.

（4） Data area

name  ——>  explain

DATA ——>  Dataset type （ Here are four data sets , By default, we select the first ; The selected data will also be displayed on the far right OUTPUT in ; In this data , We can see blue and yellow dots on the two-dimensional plane ; Each dot represents a sample example ; The color of the dot represents the label of the sample ; Because there are only two colors , So here is a binary classification problem ; Here we take the example of judging whether the parts of a factory are qualified , Then yellow represents unqualified parts , Blue represents qualified parts ）.

Ratio of training to test ——>  The proportion of data used for testing （ Directly operate the progress bar to adjust ）. Noise Introduce noise into the data .

Batch size ——>  adjustment batch size Size .

（5） Network structure adjustment area

name   ——>  explain

FEATURES  ——>  Eigenvector （ In order to map a practical problem to a point in space , We need to extract features . Here we can use the length and quality of parts to roughly describe ; So here x1 It represents the length of the part ,x2 Represents the quality of parts ; The eigenvector is the input of the neural network ）.

HIDDEN LAYERS  ——>  Hidden layer （ The neural network between input and output is called hidden layer ; Generally, the more hidden layers of neural networks, the deeper this neural network ; Here we have a hidden layer by default , This hidden layer has 4 Nodes ）.

Directly click on each icon to select Features The type of , For the operation of hidden layer , You can directly select the plus and minus signs to obtain the desired number of hidden layers and the number of neurons in each layer .

（6） Output result area

After setting the above parameters , Click Run to observe the change of output results .

If you choose a classification problem , You can see obvious boundary changes and loss In a decreasing situation , Click on show test data You can display those who are not participating in training test Data sets , Click on Discretize output You can see the result of discretization .

demonstration

Parameter setting ： Learning rate 0.03, Activation function Tanh, The regularization L1 The proportion 0.001, Question type Classification

data ： Select the first one on the top left

Network structure ： Select with two hidden layers , The first hidden layer 4 Neurons , The second hidden layer 2 Neurons

Reference from ：https://www.jianshu.com/p/95d46de63408