Machine learning with scikit-learn
•
2 likes•1,836 views
This document provides an overview of machine learning concepts and techniques using the scikit-learn library in Python. It begins with introductions to different types of machine learning problems including supervised learning tasks like classification and regression as well as unsupervised learning problems like clustering and dimensionality reduction. It then discusses common machine learning algorithms such as support vector machines, k-means clustering, random forests, and principal component analysis. The document also covers best practices for developing machine learning models including data preprocessing, evaluating model performance, and tuning hyperparameters.
Recommended
More Related Content
What's hot
What's hot (20)
Viewers also liked
Viewers also liked (20)
Similar to Machine learning with scikit-learn
Similar to Machine learning with scikit-learn (20)
Recently uploaded
Recently uploaded (20)
Machine learning with scikit-learn
- 1. Machine Learning From xkcd
- 2. Machine learning - using scikit-learn Qingkai Kong 2017-06-28 h+p://seismo.berkeley.edu/qingkaikong/
- 4. Workshop time Learningcurve Gentle Introduction • What’s ML • Types of ML ML tours • Classification • Regression • Clsutering • Dim. reduction Common Practices h+ps://github.com/qingkaikong/20170628_ML_sklearn
- 5. What is machine learning? h+ps://github.com/qingkaikong/20170628_ML_sklearn
- 6. Data examples Tunable Model Optimization algorithm Trained Model Pipeline of training a machine learning model.
- 7. Self-driving car Voice recognition … h+ps://github.com/qingkaikong/20170628_ML_sklearn
- 9. Common types of machine learning Supervised learning Unsupervised learning
- 13. Regression X Y
- 14. Regression X Y
- 15. Classification
- 16. Classification
- 18. Clustering
- 19. Clustering
- 20. Dimensionality reduction Figure from h+ps://stats.stackexchange.com/quesIons/183236/what-‐is-‐the-‐relaIon-‐between-‐k-‐means-‐clustering-‐and-‐pca
- 21. Data examples Tunable Model Optimization algorithm Trained Model Pipeline of training a machine learning model.
- 22. Representation of data Documents Images Numbers Sounds Raw data
- 23. Representation of data Documents Images Numbers Sounds Raw data n_features n_samples Feature matrix (X) Target (y) n_samples
- 24. Optimization
- 25. How we optimize? X Y A simple example
- 27. Measuring error X Y Parameter Error ω1 e1 ω2 e2
- 28. Measuring error X Y Parameter Error ω1 e1 ω2 e2 ω3 e3
- 29. Measuring error X Y Parameter Error ω1 e1 ω2 e2 ω3 e3
- 31. ω Error e Optimization Initial start Target Of course, there are many different types of cost functions and optimization algorithms, but the general idea is very similar.
- 34. Support Vector Machine Artificial Neural Network K-means Naïve Bayes Decision Tree Random Forest Ridge Regression Self-Organizing Map Logistic Regression Linear Regression Stepwise Regression Nearest Neighbor Hierarchical clustering Principal Component Analysis Gaussian Mixture Model AdaBoost Boosting Linear Discriminant Analysis Quadratic Discriminant Analysis Convolutional Neural Network Bayesian Network LASSO Ordinary Least Squares Regression Recurrent Neural Network Generative Adversarial Networks Radial Basis Function Network
- 35. Support Vector Machine Artificial Neural Network Naïve Bayes Decision Tree Ridge Regression Self-Organizing Map Logistic Regression Linear Regression Stepwise Regression Nearest Neighbor Hierarchical clustering Principal Component Analysis Gaussian Mixture Model AdaBoost Boosting Linear Discriminant Analysis Quadratic Discriminant Analysis Convolutional Neural Network Bayesian Network Ordinary Least Squares Regression Recurrent Neural Network Generative Adversarial Networks Radial Basis Function Network K-means Random Forest LASSO
- 38. Go to notebook 01
- 39. Classification Support Vector Machine Artificial Neural Network Supervised learning
- 42. A new data point
- 43. Wide margin
- 44. Narrow margin
- 45. Support Vectors
- 46. ANN
- 48. X Σ y g Input Output feature2 feature3 ω0 ω2 ω3 X – input data y – output target ωi – weights Σ – summation g – activation function Blue circle – bias feature1 ω1 1 Z = Σ = ω0x0+ω1x1+ω2x2+ω3x3+…+ωnxn y = g(ω0x0+ω1x1+ω2x2+ω3x3+…+ωnxn)
- 49. X Σ y g Input Output 1 feature2 feature3 X – input data y – output target Σ – summation g – activation function blue circle - bias feature1 Hidden1 Σ | g Hidden3 Σ | g Hidden4 Σ | g Hidden2 Σ | g 1
- 51. . . . w1 w2 wn F(eye×w1+nose×w2+…+mouth×wn) Sheldon Cooper? Intuitive Artificial Neural Network Output Input
- 52. . . . w1 w2 wn F(eye×w1+nose×w2+…+mouth×wn) Sheldon Cooper? feedback error Intuitive Artificial Neural Network Output Input
- 53. . . . w1 w2 wn F(eye×w1+nose×w2+…+mouth×wn) Sheldon Cooper? feedback error Intuitive Artificial Neural Network Output Input
- 54. . . . w1 w2 wn F(eye×w1+nose×w2+…+mouth×wn) Intuitive Artificial Neural Network Output Input
- 55. Go to notebook 02
- 57. Simple linear regression X Y y2 line: y = a + bx ŷ2 y2-ŷ2 x1 y1 Σ (yi-ŷi)2 i = 1 n Minimize:
- 58. Random Forest
- 62. Go to notebook 03
- 63. Unsupervised Principal component analysis K-means
- 66. Kmeans
- 67. Go to notebook 04
- 68. More on common practices
- 69. Machine learning is all about GeneralizaIon
- 70. Which one is better? ? ? ?
- 71. Clearly, black is better Underfitting Just right Overfitting
- 73. Test Train Train/test dataset split 20% 80%
- 77. Age Salary 24 $110,000 36 $130,000 38 $80,000 44 $420,000 27 $420,000 43 $12,000,000 … … Why do we need?
- 78. Check out: h+p://scikit-‐learn.org/stable/modules/preprocessing.html • Standardization o Zero mean and unit variance • Scale to a range o i.e. (0, 1) • Normalization o Unit norm • … Many ways
- 79. Go to notebook 05
- 80. More resources
- 82. Conclude with notebook 06