Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Gurneet1928/TryMLEasy/llms.txt

Use this file to discover all available pages before exploring further.

TryMLEasy has a public Improvement.txt file in the repository that catalogues feature ideas and known limitations gathered since the project’s initial release. This page compiles those items into a living roadmap so contributors can quickly understand where the project is headed and where their efforts will have the most impact.

​
Overall Improvements

These enhancements affect the entire application rather than a single page, and most of them would deliver a noticeably better experience for all users.
The current app uses Streamlit’s default theme. Investing in a custom color scheme, improved typography, and a more polished visual identity would make TryMLEasy feel more professional and easier to navigate. Contributions here could include a custom config.toml theme, consistent iconography, and responsive layout adjustments for smaller screens.
Several pages recompute results (such as dataset statistics or trained model outputs) on every interaction because caching is not consistently applied. Adding Streamlit’s @st.cache_data decorator to expensive operations — like reading and preprocessing a CSV or storing a trained model — would make the app noticeably faster and reduce unnecessary recomputation across page navigations.
Charts appear across multiple pages (correlation heatmaps, training/validation loss curves, etc.) but their sizing and interactivity are inconsistent. Improvements here include standardizing figure dimensions, switching to interactive Plotly charts where Matplotlib is currently used, and adding axis labels and legends that resize gracefully on different screen resolutions.
After training a model, there is currently no way to test it against a custom set of values without writing code. A final dedicated page where users can enter manual feature values and immediately receive a prediction from the trained model would complete the end-to-end no-code workflow and make TryMLEasy far more useful as a demonstration and prototyping tool.

​
Per-Page Improvements

The improvements below are scoped to individual pages in the application. Use the tabs to browse by page.
Persistent file storage in session stateCurrently, if a user navigates away from the main upload page and then returns, the uploaded dataset is lost and must be re-uploaded. Storing the uploaded file object (or its parsed DataFrame) in st.session_state with a stable key would allow the file to persist for the entire browser session, making the back-and-forth navigation between pages much smoother.
# Example: persist the dataframe in session_state on upload
if uploaded_file is not None:
    st.session_state["dataset"] = pd.read_csv(uploaded_file)

​
Known Limitations

The following are current limitations of TryMLEasy that contributors should be aware of when working on the codebase:
  • Numerical features only — all preprocessing, feature selection, and model training pipelines assume numeric input. Categorical columns must be manually encoded before upload; native encoding support is not yet implemented.
  • Manual index entry in the layer table — Streamlit’s data_editor component requires explicit index values for new rows. This is a known UX limitation; until a workaround or upstream fix is available, users must enter index numbers themselves when building a neural network architecture.
  • Single-value targets only — the output layer is always Dense(1), meaning only scalar regression or binary classification targets are supported. Multi-output and multi-class problems with more than two classes are not yet handled.
  • No model persistence between sessions — trained models (both traditional sklearn models and Keras neural networks) exist only in st.session_state and are lost when the browser tab is closed or refreshed. There is no save-to-disk or download-model feature yet.
  • No hyperparameter tuning for traditional ML models — sklearn models are trained with their default hyperparameters. There are no controls for grid search, random search, or manual parameter adjustment for any of the ten supported traditional models.

​
How to Suggest Improvements

If you have a new idea that isn’t listed here, or you’ve found a limitation not covered above, there are two easy ways to share it:
  • Open a GitHub issue — visit https://github.com/Gurneet1928/TryMLEasy/issues and describe your suggestion. Label it as enhancement if you’re requesting a feature, or bug if you’ve found a defect.
  • Email the maintainer — reach out to gurneet222@gmail.com with your idea, especially if you’d like to discuss it before investing time in an implementation.
Once you’re ready to build something, head over to the Contributing Guide for step-by-step instructions on forking the repo, setting up your branch, and submitting a pull request.
Every item on this page is an open invitation. Pick anything from the per-page improvements or overall improvements list, open a PR, and you’ll be making a real difference to TryMLEasy users. No contribution is too small — even fixing a single known limitation helps!

Build docs developers (and LLMs) love

Get started for freeTalk to us