Preprocessing presentation
0 likes307 views
Feature scaling is a data preprocessing technique that standardizes the range of independent variables and is essential for various machine learning algorithms, such as regression and neural networks. Common methods include calculating z-scores and min-max normalization. Best practices indicate that scaling should be applied only to the training set to prevent data leakage.
![● Andrew Ng
○ Okay if variables in the [-⅓ , ⅓ ] to [-3 , 3]
range
○ Otherwise, scale
When to use feature scaling?](https://image.slidesharecdn.com/preprocessingpresentation-150507202633-lva1-app6891/85/Preprocessing-presentation-5-320.jpg)
Preprocessing presentation
- 2. What is feature scaling? Feature scaling is a method used to standardize the range of independent variables or features of data. In data processing, it is also known as data normalization and is generally performed during the data preprocessing step.
- 3. ● Many estimators require it ○ Regression ○ SVM ○ kNN ○ Neural networks ● Image processing Why use feature scaling?
- 4. ● Methods ○ Calculate z-scores of your feature variables ○ Min-max method How to use feature scaling?
- 5. ● Andrew Ng ○ Okay if variables in the [-⅓ , ⅓ ] to [-3 , 3] range ○ Otherwise, scale When to use feature scaling?
- 6. ● preprocessing.scale ○ by default scales feature to mean 0 and variance 1 ○ i.e calculates the Z-score ● preprocessing.normalize ○ scales each feature to unit vector ○ useful for text classification/clustering sklearn.preprocessing
- 7. Results Pre-scaled accuracy precision recall f1 Scaled accuracy precision recall f1 LogisticReg 0.838 0.857 0.788 0.817 LogisticReg 0.835 0.850 0.787 0.813 SVMC 0.539 0.000 0.000 0.000 SVMC 0.828 0.846 0.773 0.803 GaussNB 0.828 0.828 0.802 0.811 GaussNB 0.828 0.828 0.802 0.811 DecisionTree 0.703 0.688 0.736 0.690 DecisionTree 0.714 0.720 0.729 0.693 RandomForest 0.801 0.840 0.685 0.741 RandomForest 0.818 0.845 0.700 0.773 kNN9 0.653 0.637 0.576 0.605 kNN9 0.825 0.841 0.766 0.800
- 8. Actually this is wrong
- 9. Fit scaler only on training set std_scale = preprocessing.StandardScaler().fit(X_train) X_train_std = std_scale.transform(X_train) X_test_std = std_scale.transform(X_test) also MinMaxScaler()