Supervised learning
- 2. 2 Legal Disclaimers No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document. Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software or service activation. Performance varies depending on system configuration. Check with your system manufacturer or retailer or learn more at intel.com. Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade. Copies of documents which have an order number and are referenced in this document may be obtained by calling 1-800-548-4725 or by visiting www.intel.com/design/literature.htm. Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries. *Other names and brands may be claimed as the property of others Copyright © 2018 Intel Corporation. All rights reserved.
- 3. 3 Learning Objectives • Name the steps in the data science workflow • Explain how to formulate a supervised learning problem • Compare “Training” and “Inference” • Describe the dangers of overfitting, and training vs. testing data • Identify the advantages of the PyData* ecosystem You will be able to: *Other names and brands may be claimed as the property of others.
- 5. 5 Machine Learning The study and construction of programs that learn from repeatedly seeing data, rather than being explicitly programmed by humans. Data points have known outcome Supervised Learning Unsupervised Learning Dataset Data points have unknown outcome Type
- 6. 6 Target vs. Features TargetFeatures Target: Column to predict Features: Properties of the data used for prediction (non-target columns)
- 7. Example: Supervised Learning Problem NOT SPAM? Supervised Learning Program SPAM? Goal: Predict if an email is spam or not spam. Data: Historical emails labeled as spam or not spam. Target: Spam or not spam Features: Email text, subject, time sent, etc.
- 8. Example: Supervised Learning Problem Goal: Predict location of bounding box around an object. Data: Images with bounding box locations. Target: Corners of bounding box Features: Image pixels
- 10. 10 Data Science Workflow Problem Statement Data Collection Data Exploration & Preprocessing Modeling Validation Decision Making & Deployment What problem are you trying to solve? What data do you need to solve it? How should you clean your data so your model can use it? Build a model to solve your problem? Did I solve the problem? Communicate to stakeholders or put into production?
- 11. 11 This Lesson’s Focus: Modeling and Validation Problem Statement Data Collection Data Exploration & Preprocessing Modeling Validation Decision Making & Deployment What problem are you trying to solve? What data do you need to solve it? How should you clean your data so your model can use it? Build a model to solve your problem? Did I solve the problem? Communicate to stakeholders or put into production?
- 13. Formulating a Supervised Learning Problem 13 For a Supervised Learning Problem: • Collect a labeled dataset (features and target labels). • Choose the model. • Choose an evaluation metric: “What to use to measure performance.” • Choose an optimization method:1 “How to find the model configuration that gives the best performance.” 1There are standard methods to use for different models and metrics.
- 14. Decision tree Which Model? 14 There are many models that represent the problem and make decisions in different ways each with their own advantages and disadvantages. A decision tree makes predictions by asking a series of yes/no questions. Nearest neighbor makes predictions by having the most similar examples vote. Nearest neighbors
- 15. Which Model? 15 Some considerations when choosing are: • Time needed for training • Speed in making predictions • Amount of data needed • Type of data • Problem complexity • Ability to solve a complex problem • Tendency to overcomplicate a simple one
- 16. Accuracy target Evaluation Metric 16 There are many metrics available1 to measure performance, such as: • Accuracy: how well predictions match true values. • Mean Squared Error: average square distance between prediction and true value. 1The wrong metric can be misleading or not capture the real problem. Mean square error formula
- 17. Evaluation Metric 17 Email The wrong metric can be misleading or not capture the real problem. For example: consider using accuracy for spam/not spam. • If 99 out of 100 emails are actually spam, then a model that is predicting spam every time will have 99% accuracy. • This may force an important real email into spam, even though it has a high accuracy metric.
- 18. Training 18 Training Data: The dataset used to train the model. Optimization: Configures the model for best performance.
- 19. ? Error Training 19 Forward Backward “Bicycle!” “Strawberry?” Human Bicycle Strawberry With these pieces, a model can now be trained to find the best configuration. Labeled Data Model Evaluation Metric
- 20. Inference 20 Once the model is trained, we can provide new examples for predictions. ?????? “Bicycle!” ModelNew Data Prediction
- 21. Training vs. Inference 21 Goal: Perform well on unseen data during inference. ?????? “Bicycle!” ? Error “Bicycle!” “Strawberry?” Human Bicycle Strawberry Model weights TrainingInference
- 22. Supervised Learning Overview 22 Data with answers (features and labels) Model+ Trained Model Fit Data without answers (features only) Trained Model + Predicted Answer Predict Training: Train a model with known data. Inference: Feed unseen data into trained model to make predictions.
- 24. Curve Fitting: Overfitting vs. Underfitting Example 24 X Y Goal: Fit a curve to the data.
- 25. X Y Curve Fitting: Underfitting Example 25 The curve can be too simple. • This is called “underfitting” • Poor fit on training data • Poor fit on unseen data Underfitting: Model is missing systematic trends in data.
- 26. X Y Curve Fitting: Overfitting Example 26 The curve can be too complex. • This is called “overfitting” • Good fit on training data • Poor fit on unseen data Overfitting: Model is too sensitive and fits the “noise” in the training data.
- 27. Curve Fitting Problem 27 Problem: Unseen data isn’t available during training. • How can performance be estimated? When measuring performance on the training data, there is a tendency to overfit. Underfit Well fit Overfit
- 28. 28 Solution: Split Data Into Two Sets Training Set Testing Set Training Set: Data used during the training process. Test Set: Data used to measure performance, simulating unseen data1. 1 Not used during the training process. *Other names and brands may be claimed as the property of others.
- 29. 29 Train-Test Split Model Weights Training Data Testing Data Evaluate trained model on data it hasn’t “seen” before to simulate real-world inference. “Trained Model” Training Inference Prediction
- 30. *Other names and brands may be claimed as the property of others.
- 31. 31 Why Python*? • General purpose language. • Simple, readable syntax relative to other languages, such as Java* or C++. • Has a good REPL. • Can facilitate applications written in other languages, C++ and Fortran. • Active community. • Extensive libraries. *Other names and brands may be claimed as the property of others.
- 32. 32 Python*: Fastest Growing Programming Language1 1Source: Stack Overflow *Other names and brands may be claimed as the property of others.
- 33. 33 Python*: Highest Ranked Language1 1Source: IEEE Spectrum *Other names and brands may be claimed as the property of others.
- 34. 34 Python* Libraries for Data Science Fast numerical computing Data analysis and manipulation Machine learning Visualization Deep Learning *Other names and brands may be claimed as the property of others.
- 35. Intel® Distribution for Python* Easy, Out-of-the-box Access to High Performance Python* • Prebuilt, optimized for numerical computing, data analytics, High Performance Computing (HPC). • Drop in replacement for your existing Python (no code changes required). • Download from https://software.intel.com/en-us/distribution-for-python *Other names and brands may be claimed as the property of others.
- 36. Intel® Distribution for Python* Drive Performance with Multiple Optimization Techniques • Accelerated NumPy/ SciPy / Scikit-Learn with the Intel® Math Kernel Library (Intel® MKL). • Data analytics with pyDAAL, enhanced thread scheduling with Intel® Thread Building Blocks (Intel® TBB), Jupyter* Notebook interface, Numba, Cython*. • Scale easily with optimized MPI4Py* and Jupyter* notebooks. *Other names and brands may be claimed as the property of others.
- 37. Intel® Distribution for Python* Faster Access to Latest Optimizations for Intel® Architecture • Distribution and individual optimized packages available through Conda* and Anaconda Cloud*. • Optimizations upstreamed back to main Python* trunk. *Other names and brands may be claimed as the property of others.
- 38. Train and Test Splitting: Syntax Import the train and test split function. from sklearn.model_selection import train_test_split Split the data and put 30% into the test set. train, test = train_test_split(data, test_size=0.3) Other method for splitting data. from sklearn.model_selection import ShuffleSplit
- 39. Import the class containing the classification method. from sklearn.neighbors import KNeighborsClassifier Create an instance of the class. KNN = KNeighborsClassifier(n_neighbors=3) Fit the instance on the data and then predict the expected value. KNN = KNN.fit(X_data, y_data) y_predict = KNN.predict(X_data) K Nearest Neighbors: The Syntax
- 40. 40 Learning Objectives Recap • Name the steps in the data science workflow • Explain how to formulate a supervised learning problem • Compare “Training” and “Inference” • Describe the dangers of overfitting, and training vs. testing data • Identify the advantages of the PyData* ecosystem In this lesson, we worked to: *Other names and brands may be claimed as the property of others.