This note comes from CDA-DSC Course , By Mr. Chang Guozhen . The first phase of the camp is the theme of risk control , The training content is very compact , very nice , recommend ：CDA Data scientist training camp

——————————————————————————————————

One 、logit Source of value

Logistic regression generally classifies dependent variables into two categories 0-1 Turn to frequency [0,1], become odds（ The odds ratio is ,[0,+∞]）, then log To become Logit value ([-∞,+∞]) The odds ratio is ：odds=P(y=1)/P（y=0）

logit value ：logit=log(odds)

What is? sigmoid function ？

First, I defined an intuitive concept The odds ratio is  p/(1-p),p yes true The probability of time ,1-p yes false The probability of time , Take advantage of log, namely t=log(p/(1-p)) Do a range conversion , Go to all real fields . Then turn around and ask p, In the end sigmoid function . sigmoid The interesting thing about functions is , The independent variable is from negative infinity to positive infinity , The dependent variable is 0 To 1. The closer the 0 The greater the change . The derivative is p(1-p), Derivatives are interesting .（ Reference resources ： Talk about logical regression

——————————————————————————————————

Two 、logit modeling utilize logit=Y Modeling , obtain Logit Then we can calculate the probability according to it .Logit= Economic utility , Utility is a continuous variable ,logit The model is equivalent to utility modeling .

So generally speaking , The coefficient of logistic regression is logit The coefficient of the value , It needs to be converted to a probability value .

The simple understanding can be regarded as ：

Input is x, The output is y, One of the temporary variables in the middle is t.w and b Is the model parameter .h(t) Is the probability of belonging to a certain category , Greater than 0.5 I think it belongs to this category , namely y=1.  For simplicity , We can argue that b Always with a value of 1 Of w Multiply . So we put b Put in w. The model is simplified as This is the formula of logistic regression , It's simple .

（ Reference resources ： Talk about logical regression

——————————————————————————————————

3、 ... and 、logit Function modeling threshold setting

In the risk control model summary ,logistics The threshold is set according to the business owner . Generally, high credit passes automatically , The risks in the process need to be reviewed ; The riskier refuse to borrow . ——————————————————————————————————

Four 、R Language implementation

1、 Logical regression

Logistic regression generally uses glm Function binomial(link='logit') Modeling .

summary(lg)

There are only two uses for the regression coefficient ： The sign 、 Significance . The regression coefficient represents every increase in 1 A unit of x, Will increase logit Increase in value 0.1 A unit of , And it has a positive effect . If you need to know the probability value, you need to recalculate .

2、 Stepwise regression was used to screen variables ——step

On top of logical regression , We can use stepwise regression , Eliminating variables .

lg_ms<-step(lg,direction = "both")
summary(lg_ms)

3、 Validation set prediction ——predict

train\$lg_p<-predict(lg_ms, train)
summary(train\$lg_p)

predict The same is true of the forecast for logit value , It's not probability , It needs to be recalculated

4、 Calculate the probability value

1/(1+exp(-1*train\$lg_p))

5、 Methods of model validation

As a sort class model , It can be used ROC curve /AUC value 、 Cumulative lift curve 、K-S curve 、 Lorentz curve gini To verify （ note ︱ Types of risk control classification models （ Decision making 、 Sort ） Comparison and model evaluation system （ROC/gini/KS/lift））.

note +R︱Logistics Modeling brief （logit value 、sigmoid function ） More articles about

1. note +R︱ Neural network activation function and perceptron in credit risk modeling

I always think that climbing mountains is small , can . Every time I come to the starting point , Daniel , Come slowly and share with me ,please~ --------------------------- This note comes from CDA-DSC Course , By Mr. Chang Guozhen ...

2. 【 turn 】Pandas Learning notes （ Four ） Handling missing values

Pandas Learning notes series : Pandas Learning notes ( One ) Basic introduction Pandas Learning notes ( Two ) Select data Pandas Learning notes ( 3、 ... and ) modify & add value Pandas Learning notes ( Four ) Handling missing values Pandas learn ...

3. 【 turn 】Pandas Learning notes （ 3、 ... and ） modify &amp; add value

Pandas Learning notes series : Pandas Learning notes ( One ) Basic introduction Pandas Learning notes ( Two ) Select data Pandas Learning notes ( 3、 ... and ) modify & add value Pandas Learning notes ( Four ) Handling missing values Pandas learn ...

4. R Special values in language NA NULL NaN Inf

These are R Special values in language , All are R Reserved words , NA:Not available  Indicates the missing value    use is.na() To determine whether it is a missing value NULL: Indicates null value , That is, there is no content   use is.null() To judge ...

5. c The pointer -- note 2 Function to return pointer value

Function to return pointer value Generally, functions that bring back pointer values , The definition is in the form of : int *a (int x, int y); Look at this classic case : #include <stdio.h> int main(int arg ...

6. Softmax And Sigmoid The connection of functions

7. Logistic Return to (sigmoid function , Mobile phone evaluation , The gradient rises , Batch gradient , Random gradient , A hernia disease predicts the death rate of a sick horse

( The color of the cell phone , size , User experience is used to weight the total value ) Maximum likelihood estimation MLE 1.Logistic Return to Logistic regression ( Logical regression ), It's a sort of classification , Used in dichotomous problems ( There are only two kinds of outputs ). Such as ...

8. Little pig C++ Basic notes （ 6、 ... and ） Parameter passing 、 function overloading 、 A function pointer 、 Debugging help

Little pig C++ Basic notes ( 6、 ... and ) ———— Parameter passing . function overloading . A function pointer . Debugging help key word : Parameter passing . function overloading . A function pointer . Debugging help Because of some things and my laziness , I haven't read any more for about a week , Not anymore , hurriedly ...

9. Why does logistic regression use sigmoid function

Logistic The purpose of regression is to learn one from the features 0/1 Classification model , This model takes the linear combination of characteristics as independent variables , Because the value range of independent variable is from negative infinity to positive infinity . therefore , Use logistic function ( Or called sigmoid function ) Will self ...

Random recommendation

1. Hang Dian ACM1005

Number Sequence Time Limit: 2000/1000 MS (Java/Others)    Memory Limit: 65536/32768 K (Java/Others)T ...

2. java Multithreading -- Realization Runnable Interface

package unit8; import java.applet.Applet; import java.awt.Label; import java.awt.TextField; public c ...

3. double Type into Integer type ---DecimalFormat

hypothesis x It's what you want to convert double Type variable : No rounding : (int)x Perform rounding operations : Integer.parseInt(new java.text.DecimalFormat(" ...

topic Description FatMouse prepared M pounds of cat food, ready to trade with the cats guarding the ware ...

Servlet Servlet API Overview Java Web Access to resources path problem getParameter And request.getAttribute The difference between The difference between forwarding and redirection chinese ...

6. The state of computer vision and artificial intelligence ： We've come a long way The state of Computer Vision and AI: we are really, really far away.

The picture above is funny. But for me it is also one of those examples that make me sad about the o ...

7. Interpretation of martial arts &amp;#183; scattered

If you defend your left . There must be a weakness on your right : If you defend your front , There must be a weakness behind you . If you guard carefully everywhere , There must be weaknesses everywhere ." Some martial arts are ahead of others . But it's like drinking water , The more you look, the more tasteless you feel : but ...