What it does
Domain experts often need a regression answer without writing pandas. This app:
- Accepts a CSV (or URL)
- Lets you drop columns and choose a target
- Trains a selected model (sklearn regressors or LSTM)
- Returns error metrics, a plot, and a downloadable model file
Tradeoffs worth knowing
- LSTM in the same dropdown as linear regression is powerful and easy to misuse: sequential framing is not obvious to every user.
- A single global training object means concurrent users would collide. For a multi-user deployment, per-session state or a job queue is the next step.
What it demonstrates
- Product framing of ML: hide the library surface, expose the decisions people actually make
- Composing preprocess → split → fit → metrics → viz as separate steps
- Knowing what changes between a single-user tool and a multi-user product (per-session state, a job queue)