Tutorial: A Practical Introduction To Machine Learning
summary
730 infrared grayscale bee-tag images at 24x24 pixels are loaded into numpy arrays, cropped to 16x16 (576 features) and labeled 0/1/2 for control/treatment/queen. Data splits 80/20 -> 584 train / 146 test. Visuals use PCA and LDA; a support vector machine hits ~69% accuracy, switching to a random forest raises accuracy by ~8% (≈77%). A scikit-learn pipeline wraps an unwrap transformer + LDA + classifier, and hyperparameters are tuned with RandomizedSearchCV; improper testing on training shows a spurious 96% but proper held-out testing yields ~78%, confirming the pipeline works and is reproducible.
AI-written summary · about
chapters (12)
- 0:00Start
- 19:37crop the image to the central square
- 26:03load images in as numpy
- 41:07plot the data on the x and y axis
- 42:51try and visualize with another type of algorithm linear discriminant analysis
- 47:22fit the classifier to the training data
- 48:33predicting the label of each of the testing rows
- 48:48use a very simple measure of accuracy
- 49:51baseline accuracy measure
- 1:06:16build a pipeline
- 1:13:25using the randomized search cross-validation
- 1:18:08repeat this process using a proper training test split
related talks
24mWhat's the Buzz with Machine Learning
36mDetecting Honeybee Swarms Using the Integration of OpenCV, Pandas, AI, & PyTorch
35mIs it Food? An Introduction to Machine Learning
23mBirding with Python and Machine Learning
40mMachine learning 101: Where to begin?
45mApplied machine learning: a few lessons I learned the hard way
43mLessons Learned Using Machine Learning in Java
01h 17mMachine Learning with Scikit-Learn