Eye PyTorch training camp depth of the second period --- 4, autograd and logistic regression

An automatic derivation system ---- torch.autograd

1, torch.autograd.backward  functions: automatic strike gradient

• • 　　tensors: tensor for the derivation of such loss
  •        retain_graph: Save the calculation of FIG.
  •        create_graph: Create FIG derivative calculation for higher order derivative
  •        grad_tensors: multi-gradient weights

2, torch.aurograd.grad  function: to strike gradient

• • 　　outputs: Tensor for derivation of Y -----
  •         inputs: gradient tensor requires ----- w, x
  •         create_graph: Create FIG derivative calculation for higher order derivative
  •         retain_graph: Save the calculation of FIG.
  •         grad_outputs: multi-gradient weights

Tips:

　　(1) is not automatically cleared gradient  grad.zero_ ()

　　(2) dependent on the leaf node of the node, requires_grad defaults to True

　　(3) non-leaf node performs in-place operation

　　　　

Second, the logistic regression

Logistic regression model is linear dichotomous

Model expression: y = f (WX + b)

                      f (x) = 1 / (1 + e (-x)) f (x) called the Sigmoid function, also known as Logistic Function

　　　　　　class={0，0.5>y

　　　　　　　　　1，0.5<y

• The difference between the linear regression and logistic regression
  • Linear regression analysis is the independent variable x and (y dependent variable scalar ) Method The relationship between
  • Logistic regression analysis is the independent variable x and dependent variable y ( probabilistic ) method of the relationship between

 

• Step training machine learning models
  • 　　data
  •        model
  •        Loss function
  •        Optimizer
  •        Iterative training