Tensors from scratch

From the Python list you already know how to write to a fraud detector that learns from data.

Who it is for: Computer Systems students who already program a bit in Python, and Business Administration people who want to understand what lies beneath the artificial intelligence used in banks, insurance companies and stores. It is an intermediate-level course: it does not explain what a variable or a loop is, but every new word is explained when it appears. Each session includes an “If you come from Business Administration” box with the question worth asking in a meeting.

What you will know by the end:

• Create a PyTorch tensor from real data and state its shape before running the code.
• Extract and summarize data along any of its axes, and check each result against the same calculation done with a for.
• Read a PyTorch shape error, word by word, and fix it.
• Train a small fraud detector and explain why getting 91 % right is not enough.

What you need before you start: basic Python (variables, lists, for loops and functions; lessons 0 to 4 of the Python course are enough), Python 3.11 or newer and a terminal. The first session installs nothing; PyTorch is installed in the second. Tested on Linux (Mint 22.3 and Ubuntu 24.04) and on macOS; it has not been tested on Windows. No graphics card needed.

The eight sessions

Two lessons of four sessions, about 2 hours each: roughly one hour to read and follow the example, and another to practice. They all follow the same story: a stationery store, a card terminal that overcharges, and a student who has until December 23 to understand what happened.

Session		What you learn
1	A number, a list, a table, a cube	what a tensor is, with Python lists; counting dimensions; where the word came from
2	Your first tensor	installing PyTorch; torch.tensor, shape and dtype; PyTorch and TensorFlow
3	Axes have names	taking a slice along any axis; what happens to the shape
4	Summing by tab	summing and averaging over an axis with dim
5	Changing shape without changing the data	reshape and how to read a shape error
6	Operating on full tensors	broadcasting, and the trap of * versus @
7	A fraud detector that learns from data	the training loop; accuracy, recall and precision
8	Tensors in the bank, the insurance company and the store	reading a real case and stating the shape of its tensor; the precautions that are not optional

The code

Each session has its folder in this repository (sesion-01 to sesion-08), with every program and the output it should give you. verificar.py runs all of them and compares them with what the lesson says. The step-by-step installation is in the main README.

Where it is published

One session per week on “En construcción”, Hábil’s Substack, on Wednesdays at 8:00, from November 4 to December 23, 2026: https://enconstruccionmx.substack.com