LearnAISign in
Phase 2 · 5 lessons · 15 h · about 1.5 weeks

⌘ Programming & Tooling for ML

The minimum stack that lets you run every idea in this path

Lessons

2.1Python for ML, fastThe 20% of the language that writes 95% of ML code2 h 30 min2.2NumPy and vectorized thinkingShapes, broadcasting and the end of for-loops3 h2.3pandas, data wrangling and plottingGetting a messy table into a shape a model can learn from3 h2.4PyTorch I: tensors, autograd and the training loopThe five lines every deep-learning program is built from3 h 30 min2.5Tooling: Git, GPUs, experiments and the Hugging Face HubMaking your work reproducible, shareable and affordable2 h 30 min

About this phase

Every later phase assumes you can turn an equation into working code in minutes: load a dataset, reshape it into tensors, write a training loop, plot what happened, and save the result somewhere you can find again. This short phase builds exactly that muscle and nothing more. You will learn the slice of Python that machine-learning code actually uses, think in arrays with NumPy (where broadcasting and vectorization replace loops), wrangle and plot data with pandas and matplotlib, meet PyTorch's tensors and autograd, and set up the tooling — environments, Git, GPUs, experiment tracking, the Hugging Face Hub — that keeps projects reproducible and cheap. If you already code daily, skim the notes and do the practice tasks; they are calibrated so that finishing them means you are ready for Phase 3.

By the end you can

  • Write idiomatic Python for data work: comprehensions, dataclasses, generators, typing, pathlib, and isolated environments with uv
  • Reason about array shapes, broadcasting and vectorization well enough to implement softmax, pairwise distances and a linear layer without loops
  • Load, clean, group, join and visualize a real dataset with pandas, matplotlib and seaborn, following an EDA checklist
  • Explain what autograd computes and write the canonical PyTorch training loop from memory
  • Run experiments reproducibly: Git workflow, free GPUs on Colab/Kaggle, seeds, experiment tracking and the Hugging Face Hub