Baseline Model Plan Before Complexity
Build the simplest model that could work.
The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.
Plan a baseline for my machine learning task. Task and target: [TASK_AND_TARGET]. Data: [DATA_DESCRIPTION]. Time budget: [TIME_BUDGET]. Language and libraries: [LANGUAGE_AND_LIBRARIES]. Produce a staged plan: 1. Dumb baselines: predict the majority class or the mean, and last-period value for time series. Record their scores. 2. Rule baseline: one or two rules a domain expert would write. 3. Simple model: logistic or linear regression with regularization, with feature handling (scaling, one-hot encoding, missing indicators). 4. Tree-based model as the first non-linear step. 5. A table to track runs: Run | Data version | Features | Model | Parameters | Metric | Notes. 6. A stopping rule: stop adding complexity when the gain on validation is below a threshold I set or when error analysis shows the problem is the data, not the model. 7. Error analysis: look at 50 worst predictions and classify why they failed. Provide the code skeleton with a pipeline and cross-validation. Explain why a baseline protects me from fooling myself.
Fill in before use
Optional: type your details and the prompt above updates. Nothing is saved or sent.
Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.