Next Steps

Using this seminar to improve your research

Everything in the labs ran inside claude.ai in your browser: you upload a file, ask in plain English, and Claude computes the answer. That is perfect for learning and for small-to-medium datasets. But it has real limits — and the good news is that the same skill you practiced lifts those limits.

Where the browser reaches its limit

• File size. Very large datasets (hundreds of thousands or millions of rows, or files above the upload limit) are too big to attach to a chat.
• Repeatability. Running the same analysis over 50 files, or re-running it monthly, is tedious by hand in a chat.
• Heavy computation. Training large models, deep neural networks, or big text models needs more memory and processing power than a browser session provides.

What you install once (all free)

You need two things: a language to run the code, and an editor to write and run it in. Pick one language and one editor.

The new prompt: ask for code, not answers

The prompt is almost identical to the ones you used in the labs — you just ask for a script and point it at a folder on your computer instead of an uploaded file. Claude writes the complete program; you paste it into your editor and press Run.

Write a complete Python script that:
1. Reads my data file "GOV2.csv" from the folder C:\data\ (columns: country, iso3, year, Y1, Y2, X1-X20).
2. Builds a classification tree (CART) predicting Y2 from X1-X20 only, with a 70/30 train/test split and max depth 3.
3. Prints the tree rules as text, the test accuracy, and the confusion matrix.
4. Uses only common libraries (pandas, scikit-learn) and includes comments explaining each step so I can learn from it.
Tell me exactly which libraries to install first and the command to run the script.

Swap the details for any analysis from the seminar — random forest, neural network, dictionary sentiment, LLM-style text extraction — and for R, just say “Write an R script that…” instead. Claude will also tell you the one-time install command (for example pip install pandas scikit-learn) and how to run the file.

How it works, start to finish

If a script errors, copy the red error message back into Claude and ask “how do I fix this?” — debugging by pasting the error is a normal, fast part of the workflow, and a great way to learn.

Very large data & when you need a GPU

Your laptop's regular processor (the CPU) handles the trees, forests, and small models from this seminar comfortably, even on large tables. Some tasks, though, are dramatically faster on a GPU — a graphics processor built for the massive parallel math that deep learning uses.

• CPU is fine for: CART, random forests, regressions, dictionary methods, and most tabular data — even millions of rows.
• A GPU helps for: training deep neural networks, fine-tuning large language or image models, and heavy text/embedding work on big corpora.

You do not need to buy one. Free cloud notebooks give you a GPU in the browser: Google Colab and Kaggle Notebooks both offer free GPU time — you can ask Claude to write code specifically for Colab. For sustained or very heavy work, paid cloud GPUs (Google Cloud, AWS, Azure) or a university computing cluster are the next step.

References & free resources

A full reference deck accompanies the seminar — every paper, dataset, video, and tool, each with a free and legal way to reach it (no university subscription required). Download the complete document, or use the shortlist of essentials below.