Author | Wang Wenqi
Zebian | Carol
Source | CSDN blog
Produced | AI technology base camp (ID: rgznai100)
learning target
We know the logistic regression function loss
Optimization of logistic regression to know
We know sigmoid function
You know the scenario logistic regression
Application LogisticRegression realize logistic regression forecast
We know the difference between precision, recall indicators
We know how to solve assess imbalances in the sample
ROC curve to understand the meaning of instructions AUC index size
Classification_report precise application, recall calculated
Indicators for computing applications roc_auc_score
Logistic regression Introduction
Logistic Regression (Logistic Regression) is a machine learning classification model, logistic regression is a classification algorithm, although the name with the return, but there is some connection between it and return. Due to the simple and efficient algorithm is widely used in practice.
Logistic regression application scenarios:
CTR
It is spam
Whether sick
Financial Fraud
False account
See the example above, we can see the characteristics of them, that is all part of the judgment between the two categories. Logistic regression weapon is to solve binary classification problem.
Logistic regression principle
To master the logistic regression, we must grasp two things:
Logistic regression, what is the input value
How to determine the output logic regression
1, input
Input logic is the result of a regression linear regression.
2, activation function
sigmoid function
Criteria
The regression results are input to the sigmoid function among
Output Results: [0, 1] value of a probability interval, the default threshold value is 0.5
The final classification by logistic regression determines a probability value is belonging to a category belong to a category, and the category default flag is 1 (positive patients), another category are labeled 0 (counterexample). (Loss calculation convenience)
Interpretation of results output (important): Suppose there are two categories of A, B, and we assumed probability value belonging to A (1) the probability value of this category. Now there is a sample of the input to the output of the logistic regression 0.6, then the probability value exceeds 0.5, means that we train or predict the result is A (1) category. Conversely then, then 0.3, or training if the outcome is predicted to result of B (0) category.
So linear regression to predict the results before then we remember we use to measure the mean square error, that if for logistic regression, we predict that the results do not how to measure the losses? We look at such a picture.
So predictions of how to measure the results of logistic regression of the real difference?
Losses and optimization
1 loss
Loss of logistic regression, called log likelihood loss , the following formula:
Separate categories:
How to understand a single equation it? This should be understood in terms of image log function
Integrated complete loss of function
See this formula, in fact, similar to tell us information entropy.
Then we bring the example above is calculated again, will be able to understand the meaning.
We already know, log§, the larger the P value, the smaller the result, so we can look at this formula to analyze loss
2, optimization
Also using a gradient descent optimization algorithm to reduce the value of the loss function. Such logic to update the previous corresponding weight regression algorithm parameters of weight, enhance the probability that originally belonged to a category, reducing the probability of 0 was originally categories.
【end】
◆
Force plans
◆
"Force plan [the second quarter] - learning ability Challenge" started! From now until March 21, must flow to support the original author! Exclusive [more] medal waiting for you to challenge
Recommended Reading
How to say goodbye and hares are null? Great God turned out to be doing!
Overheating To live with the goods, Jingdong say you can help me
Your point of each "look", I seriously as the AI