Build your free machine learning for free path

We skipped Kaggle's Python micro-course (7 hours). It is built to get you into Kaggle's Pandas and machine learning micro-courses quickly, so it runs through syntax, loops, dictionaries, and external libraries and stops there. CS50P takes 36 hours and spends them on exceptions, unit tests, and file I/O as well, which is what you reach for the first time a training script dies on row 40,000 of your data.

We skipped Kaggle's Intro to Machine Learning (3 hours). It is a fast, clean run at decision trees, model validation, overfitting, and random forests in scikit-learn, and three hours of browser notebooks really is enough to enter a competition. Google's Machine Learning Crash Course takes 15 hours over the same fundamentals plus the TensorFlow APIs and worked case studies, which is the groundwork the deep learning steps below assume you already have.

We skipped Kaggle's Intermediate Machine Learning (4 hours). Its four hours go on missing values, categorical variables, pipelines, cross-validation, and XGBoost, which are the things that really do break a model on messy data. It is advice rather than practice, though. freeCodeCamp's Machine Learning with Python runs 30 hours and has you build and train five certification projects in TensorFlow, including natural language and reinforcement learning work.

We skipped Kaggle's Intro to Deep Learning (4 hours). It gets you a working Keras model fast, with dropout and batch normalization on real datasets, so you will have built something by lunchtime. What four hours cannot do is explain why any of it works, and that is this step's whole job. MIT's course is 30 hours on CNNs, RNNs, generative models, and responsible AI, with lecture videos refreshed every year.