EasyML

A small domain-specific language and Python interpreter for loading tabular data, basic cleaning, and training a choice of two sklearn models—built as a final project for a programming class.

What it is

EasyML runs .ezml scripts line by line: set a file path, load CSV or Excel, clean rows, optionally fit classification or regression (picking between two fixed model pairs), and export datasets or models to exported/. It is a learning exercise, not a production ML system.

Tech stack

Python 3.11, pandas, scikit-learn (linear/logistic regression, decision tree classifier, random forest regressor), joblib, openpyxl (Excel). Dependencies are pinned in environment.yml and requirements.txt.

Prerequisites

• Miniconda or Anaconda with conda on your PATH
• Your own CSV/XLSX data if you run the samples (see trainingData/README.md)

Setup

1. Clone this repository and cd into the project root (the directory that contains easyML.py and environment.yml).
2. Create the environment:

   conda env create -f environment.yml

3. Activate it:

   conda activate undergrad-archive--easyml

4. Create an output directory (first DOWNLOAD will fail if it is missing):

   mkdir -p exported

Usage

conda activate undergrad-archive--easyml
cd /path/to/EasyML
python easyML.py your_script.ezml

DATAPATH myPath = 'data/myfile.csv'
DATASET myDf = LOAD myPath
CLEAN myDf
MODEL myModel = PREDICT_CAT(myDf, COLUMN J)
DOWNLOAD DATASET myDf
DOWNLOAD MODEL myModel

DATAPATH myPath = 'data/myfile.csv'
DATASET myDf = LOAD myPath
CLEAN myDf
MODEL myModel = PREDICT_NUM(myDf, COLUMN I)
DOWNLOAD DATASET myDf
DOWNLOAD MODEL myModel

Sample scripts

• sample1.ezml — Loads running.xlsx, cleans, exports the dataset only.
• sample2.ezml — Loads Titanic.csv, cleans, trains classification on COLUMN J, exports dataset and model.
• sample3.ezml — Loads housing.csv, cleans, trains regression on COLUMN I, exports dataset and model.

Known limitations

• One statement per line; whitespace splitting (line.split()). Targets use COLUMN letters (A=0, …). PREDICT_CAT vs PREDICT_NUM; paths with spaces are fragile.
• No unit tests or CI; little error handling beyond a missing script file.
• No datasets in the repo; samples assume files exist at the given DATAPATH (see trainingData/README.md for formats and public sources).
• exported/ is not created automatically.
• Model choice and metrics are simplistic; multiclass and messy real-world tables can still break edge cases.

License

Apache License 2.0 — see LICENSE.