Machine Learning 101

Presentatie gegeven tijdens devCampNoord '24 in Kinepolis Groningen.

Transcript

1. Lutske de Leeuw Groningen 04-04-2024 Where to begin? Machine Learning 101
2. So many ideas
3. The original problem
4. Table of content • About me • Introduction to Machine Learning • Types of Machine Learning • Understand the problem • Data Collection and Preprocessing • Model selection and training • Model evaluation • Real life examples • Conclusion
5. https://www.linkedin.com/in/lutske/
6. Introduction to Machine Learning
7. Machine Learning ‘Teach’ the computer Without programming
8. Artificial Intelligence Imitate the human brain Including machine learning
9. Deep Learning Neural networks Subset of machine learning
10. Generative AI Generate content Subset of deep learning
11. Artificial Intelligence Machine learning Deep learning Generative AI
12. Data Science Knowledge extracted from data Advice & predict
13. Data science Artificial Intelligence Machine learning Deep learning Generative AI Data Science
14. Types of Machine Learning
15. Types of learning Supervised Unsupervised Semi-supervised Reinforcement
16. Supervised Prediction Input Label: Cat Model ? It’s a cat Label: Not a Cat
17. Unsupervised Input Model
18. Semi-supervised Input Model It’s a cat Duck Cat ? Prediction
19. Reinforcement learning Agent Environment ? Walk Sit Action= Sit Reward! Update policy Sit = good!
20. Types of data Numerical Categorial Text data Yes / No Ordered series Blood groups Measured values Sentiment analysis Translations Documents
21. Understanding the problem
22. Understanding the problem • Nature of the problem • Type of data available • Desired outcome
23. Example • Problem: My 2 cats needs different food • Type of data: lots of cat pictures • Desired outcome: Recognize cat for access to the correct bowl of food
24. Is machine learning the way? • Big if else statement? • Do I have enough data? • Is there an expert to check the data?
25. Code Example pip install pandas numpy matplotlib scikit-learn jupyter import pandas as pd import numpy as np import matplotlib.pyplot as plt from sklearn.model_selection import train_test_split from sklearn.preprocessing import StandardScaler from sklearn.linear_model import LinearRegression from sklearn.metrics import mean_squared_error
26. Data collection and preprocessing
27. Data collection • Identify data sources (databases, API’s, web scraping, manual) • Determine the data requirements (type, size, format) • Collect the data • Clean the data • Validate the data • Organize the data
28. Don’t reinvent the wheel Kaggle UCI Google
29. Don’t reinvent the wheel OpenML Data.gov Data.overheid.nl
30. Code example: Load dataset from sklearn.datasets import fetch_california_housing california = fetch_california_housing(as_frame=True) dataFrame = pd.DataFrame(california.data, columns=california.feature_names) dataFrame['prices'] = california.target
31. Code example: See the data dataFrame.head() # MedInc HouseAge AveRooms AveBedrms Population AveOccup Latitude Longitude Prices 0 8.3252 41.0 6.984127 1.023810 322.0 2.555556 37.88 -122.23 4.526 1 8.3014 21.0 6.238137 0.971880 2401.0 2.109842 37.86 -122.22 3.585 2 7.2574 52.0 8.288136 1.073446 496.0 2.802260 37.85 -122.24 3.521 3 5.6431 52.0 5.817352 1.073059 558.0 2.547945 37.85 -122.25 3.413 4 3.8462 52.0 6.281853 1.081081 565.0 2.181467 37.85 -122.25 3.422
32. Preprocessing • Data cleaning • Data transformation • Feature extraction • Feature engineering • Data splitting
33. Code example: Prepare data housing = dataFrame.drop('prices', axis=1) pricing = dataFrame['prices']
34. Data processing
35. Gather data • From your resources • From existing resources • Data Augmentation
36. Split the data Dataset Train Validation Test
37. Split the data Train 80% Test 10% Validate 10% Train 70% Test 15% Validate 15% Train 60% Test 20% Validate 20%
38. Code example: Split & scale the data housing_train, housing_test, pricing_train, pricing_test = train_test_split(housing, pricing, test_size=0.2, random_state=42) scaler = StandardScaler() housing_train_scaled = scaler.fit_transform(housing_train) housing_test_scaled = scaler.transform(housing_test)
39. Algorithms
40. Linear Regression
41. Logistic Regression
42. Decision Tree Is this person fit? Age < 30 ? Eats a lot of pizzas? Exercises in the morning? Unfit Fit Fit Unfit Yes? Yes? Yes? No? No? No?
43. Code example: Train the model & predict model = LinearRegression() model.fit(housing_train_scaled, pricing_train) pricing_pred = model.predict(housing_test_scaled)
44. Model evaluation
45. Use metrics www.mathworks.com/discovery/overfitting.html
46. Confusion Matrix Positive (Cat) Negative (Dog) True Positive True Negative False Negative False Positive Cat Cat Dog Not a cat Positive (Cat) Negative (Dog) Actual Predicted
47. Accuracy • True Positive (TP) • True Negative (TN) • False Positive (FP) • False Negative (FN)
48. Formulas
49. Mean Squared Error
50. Mean Squared Error
51. Code example: Mean Squared error mse = mean_squared_error(pricing_test, pricing_pred) print(f"Mean Squared Error: {mse:.2f}") Mean Squared Error: 0.56
52. Improving the model
53. Improving the model • Feature Engineering • Hyperparameter tuning • Model selection • Handling outliers • Cross Validation • Collect more data • Regularization
54. Code example: Handling Outliers # Example: Winsorization from scipy.stats.mstats import winsorize dataFrame['prices'] = winsorize(dataFrame['prices'], limits=[0.05, 0.05])
55. Code example: Using another model # Example: Descision tree from sklearn.tree import DecisionTreeRegressor model = DecisionTreeRegressor() model.fit(housing_train_scaled, pricing_train) pricing_pred = model.predict(housing_test_scaled) mse = mean_squared_error(pricing_test, pricing_pred) print(f"Mean Squared Error: {mse:.2f}") Mean Squared Error: 0.49
56. Code example: Using another model # Example: Random Forest Regression from sklearn.ensemble import RandomForestRegressor model = RandomForestRegressor() model.fit(housing_train_scaled, pricing_train) pricing_pred = model.predict(housing_test_scaled) mse = mean_squared_error(pricing_test, pricing_pred) print(f"Mean Squared Error: {mse:.2f}") Mean Squared Error: 0.25
57. Do try this at home! https://tinyurl.com/485n4byj https://github.com/Lutske/ML101_where_to_begin
58. Real life examples
59. Real life examples
60. Conclusion
61. Key take aways • Collect a lot of data • Don’t invent everything yourself • Machine learning isn’t always the solution • Verify the model
62. Questions?
63. Feedback
64. Thank you!