EasyResearch 0.1.2

EasyResearch 0.1.2 (Python package)

• Signal classification: one model with a single 80/20 validation split no longer crashes while saving predictions.csv (the file now lists the 20% validation recordings, with their rows in the data file).
• Forecasting: a date column in which some dates appear more than once is no longer refused as "not evenly spaced"; the spacing is read from the distinct dates, and repeated time points are averaged as before (with a note).
• ROCKET works on recordings of 10 values, the shortest accepted: a kernel longer than the recording is padded so it fits.

EasyResearch Desktop 0.1.2 (Windows)

• Fixed (Signal classification): one model with Single 80/20 validation split crashed while saving the results.
• Fixed (Time-series forecasting): files with repeated dates could be refused as "not evenly spaced".
• Fixed (Signal classification): ROCKET failed on recordings of 10 values, although 10 values are accepted.
• publish.ps1 starts from an empty output folder and checks that every bundled package matches requirements.lock.txt. Before, a second local build kept the packages of the first one; the setup program built by GitHub was not affected.
• Uses EasyResearch 0.1.2.

Install on Windows: download EasyResearchDesktop-0.1.2-setup.exe below and run it (no Python or administrator rights needed).
Python: pip install --upgrade easyresearch

v0.1.1: Version 0.1.1: the clean-PC fix for EasyResearch Desktop

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Azautzcb4KbHpwD3zvBJLb

EasyResearch 0.1.0

First release of EasyResearch: machine learning without programming for regression, time-series forecasting and signal classification (and EasyClassifier's classification), as a Windows desktop application and a Python package.

Install

• Windows: download EasyResearchDesktop-0.1.0-setup.exe below and run it (no Python needed).
• Python: pip install easyresearch (pip install "easyresearch[full]" adds XGBoost and LightGBM).

Highlights

• Leakage-free preparation, default parameters with fixed seeds, and an honest final score for the selected model
• Forecasting evaluated in time order, with a final period never used to choose the model
• Plain-language reports that cite every method used
• Advanced options for figures, measures, KNN distance and data preparation
• Requires EasyClassifier 0.8.2

Full details: CHANGELOG.md and desktop/CHANGELOG.md.