yibelang 2.1.0

Y-ibe - machine learning in plain English. No syntax. No cloud. Your data never leaves your computer.

Y-ibe — machine learning in plain English

No syntax. No setup headaches. No cloud. Your data never leaves your computer.

Y-ibe is a tiny programming language where you write machine learning as ordinary English sentences, one per line. The yibe tool translates your sentences into real Python (pandas + scikit-learn) and runs it locally. It is 100% deterministic — no AI, no API keys, no network calls, same input → same output, forever.

get data from "houses.csv" and call it houses
put columns of houses named square_feet and bedrooms into features
put column of houses named market_value into targets
make a linear regression model called price calculator
train price calculator using features and targets
check how good price calculator is on features and targets

$ yibe run prices.yibe
Score for price_calculator: 0.973

Six sentences. A real, trained, scored model — running the same C-powered libraries the experts use.

Install

pipx install yibelang        # recommended (or: pip install yibelang)
yibe version

More options (venv, from source) in INSTALL.md.

Try it in 60 seconds

yibe examples                          # list the bundled example scripts
yibe run examples/04_house_prices.yibe # train your first model
yibe check examples/04_house_prices.yibe   # validate without running
yibe show-python examples/04_house_prices.yibe  # see the Python it writes

show-python is the graduation path: when you're ready to learn Python, Y-ibe shows you exactly what it has been writing for you all along.

What you can say

The full sentence list lives in docs/LANGUAGE.md (it is generated from the grammar itself, so it can never go stale). Highlights:

You write	Y-ibe does
get data from "sales.csv" and call it sales	loads a CSV (or .xlsx) with pandas
remove rows with missing values from sales	cleans your table
keep rows of sales where units is greater than 0	filters rows
make a random forest model called sorter	linear/logistic regression, decision tree, random forest, neural network
split features and targets into training and testing parts	honest train/test split
train sorter using x_train and y_train	fits the model
check how good sorter is on x_test and y_test	scores it
use sorter to predict from features and call it guesses	predictions
save sorter to "sorter.yibemodel" / load model from ...	reuse models later
draw a scatter of square_feet versus market_value from houses	charts
if price limit is greater than 200000 + indented block, otherwise, repeat 3 times	decisions and loops (v1.1)
कीमत गणक को features और targets से सिखाओ · entrena calculador de precios con features y targets · 用features和targets训练价格计算器	whole other human languages — Hindi, Spanish, and Mandarin ship built in (lang: hi/es/zh); a language pack is just a folder of JSON

Errors that talk like a person

Y-ibe never shows you a raw traceback. Mistakes are caught before your script runs whenever possible, and every runtime failure is translated:

🌊 Y-ibe hit a wave:
Hey, I tried looking for your file, but it isn't in this folder.
Fix: double-check the spelling inside your quotes and make sure the file is in
the same folder you're running from. (I looked for: houses.csv)
This happened around line 1 of your script: get data from "houses.csv" and call it houses

Typos get suggestions ("Did you mean...?"), using a model before training it is caught at check time, and --debug reveals the real traceback when you want it. The full catalogue is in docs/ERRORS.md.

Honest positioning

Y-ibe is a deterministic, friendly front-end over pandas and scikit-learn — "just a DSL", and proudly so. The bet is that what actually blocks beginners is not the math, it's the syntax and the terrifying error messages. Y-ibe removes exactly those two things and nothing else:

• It is not an AI product. Fixed sentence templates, matched offline. No hallucinations, no API costs, no privacy risk.
• It is not a Python replacement. It writes Python for you, and shows it to you on request.
• It is as fast as Python + scikit-learn — because that is literally what it runs.

For teachers

Students express intent ("train the model, check it on unseen data") before they ever fight a bracket. yibe check explains mistakes in sentences, and yibe show-python turns every script into a Python lesson. The bundled examples (yibe examples) are a ready-made first lab.

Contributing

The perfect first contribution is one new sentence: a regex template + an AST node + its codegen + six tests + a doc example. See CONTRIBUTING.md. Feature ideas go through YEPs (Y-ibe Enhancement Proposals) — open an issue.

License

MIT — see LICENSE.