Hands on data analysis data modeling and model evaluation

Data modeling and evaluation

Introduce

After data processing and Preliminary visual analysis , We can use the data to get the information we want . The first step of data analysis is modeling , After modeling, we need to evaluate whether our model is reliable .

Data modeling

The modeling library used here is sklearn, It contains many algorithms of machine learning , The corresponding model algorithm selection path can refer to the following figure ：

Divide the data set

First, the data set should be divided into training set and test set , What we use here is sklearn.model_selection.train_test_split Method , Can pass jupyter Of train_test_split? View the documentation for the method .

Note that random selection is used by default for cutting data sets , It needs to be judged according to the actual situation .

Model creation

stay sklearn in , All estimators are inherited from estimator, All pass fit Method to build the model , Use predict To predict the outcome .

For classification , You can use logistic regression or random forest , Corresponding to the following two classes ：

• sklearn.liner_model.LogisticRegression
• sklearn.ensemble.RandomForestClassifier

Model to predict

After building the model , Can pass predict Method to predict the model , Input eigenvalue x, The corresponding label will be given y value .

You can also use predict_proba To get the probability of each tag corresponding to the model prediction .

Evaluation of the model

Cross validation

sklearn.model_selection.cross_val_score(estimator, X_train, y_train, cv=10)： Output the score of each cross validation

Confusion matrix and corresponding probability calculation

• sklearn.metrics.confusion_matrix
• sklearn.metrics.classification_report

draw ROC curve

sklearn.metrics.roc_curve, The return value is false positive rate、true positive rate and thresholds