➡️ Part 6 of learning ML code from scratch Your spam filter, a property site and your music app all…
MachineLearningToGo
Visit on TikTok29K followers
12,900 views #machinelearning #python #coding
➡️ Part 6 of learning ML code from scratch Your spam filter, a property site and your music app all run on the code from parts 1 to 3. Machine Learning from Scratch, part 6: real life jobs. The learning code stays exactly the same for all three examples. We only hand it different data, and each time it becomes a different product. The whole loop from parts 1 to 3 sits in one function, learn(): examples in, guess, error, small step, again and again. Sixteen lines of Python call it three times. Job 1, part 1: predict a number. Seven flats with size and price, made up for this video. The line settles, and a 90 m2 flat gets 277 thousand. Property sites do this with hundreds of details instead of one. Job 2, part 2: yes or no. Seven mails, and we count two words in each: free and meeting. The answers are yes or no, so the squashing curve goes on top, and the neuron learns one weight per word: free +5.4, meeting -5.7. A new mail that says free twice: 97.8 % spam, straight to the junk folder. Job 3, part 3: songs you liked or skipped, by tempo and energy. Your taste is a zone in the middle, and a single straight line only gets 79 % right, by saying no to almost everything. Six hidden neurons each draw a line, the error flows backwards to them (that is backpropagation), and together they fence your taste in: 97 %. A new song inside the fence: 99 %, recommended. Part 4, before anything goes live: every model is graded on examples it never saw. The hidden flats are off by 1.5 thousand, both hidden mails are right, and 18 of 21 hidden songs. Same sixteen lines, three products. Only the data changed. With billions of dials the same loop reads faces and writes text. Next time we dive deeper into activation functions, the curves that make a neuron a neuron. Make sure to follow. #machinelearning #python #coding #ai #datascience
Comments
Comments from the original post on TikTok
No comments yet
