Machine Learning logistic regression

1.sigmoid function:     logistic regression and linear regression function of contact


  The compressed G z (z), between the (0,1), a collation 0.5 or more and 0.5 or less normalized class 0

2. logistic regression Features: good fit for linear effect, fast calculation, classification results are not fixed 0 and 1, but the probability of class numbers presented in decimal form.

3. Solving loss function: maximum likelihood estimate  

4. prevent overfitting: L1 regularization    

         L2 regularization    

  C control over the degree of regularization parameters, L1 parameter n will be compressed to 0, L2 parameters will be as small as only n

5. Regulation parameters loss function: penalty regular default L2

          The default C 1, C is smaller stronger effect regularization

6. gradient descent important parameters: maximum number of iterations max_iter

7. Duality and multiple regression important parameters: solver selection method, the default liblinear coordinate descent, dichotomous

               multi_class inform the type of treatment classification model, the default ovr binary, multinomial also fill multiple classifiers, auto automatic selection 







Guess you like