> ## Documentation Index
> Fetch the complete documentation index at: https://sourcetable.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Machine learning

> Run predictions and classifications on your spreadsheet data with TabPFN and scikit-learn.

Sourcetable brings machine learning into the spreadsheet. No data science experience required — describe what you want to predict or classify, and the AI builds and runs the model for you.

## TabPFN (in-context learning)

TabPFN is a neural network that can make predictions on tabular data without training. It works directly on your spreadsheet data:

### Predict mode

Forecast numerical values based on patterns in your data:

* "Predict next month's revenue based on the last 24 months"
* "Estimate the sale price for these properties based on size, location, and age"
* "Forecast customer lifetime value based on their first 90 days of activity"

Select **Predict** mode from the AI chat mode selector, or ask in natural language.

### Classify mode

Categorize rows based on features in your data:

* "Classify these leads as hot, warm, or cold based on their engagement metrics"
* "Predict which customers are likely to churn based on usage patterns"
* "Categorize these support tickets by priority level"

Select **Classify** mode from the AI chat mode selector.

## scikit-learn

For more traditional machine learning workflows, the AI uses scikit-learn under the hood:

### Clustering

* "Group these customers into 5 segments based on their purchase behavior"
* "Find natural clusters in this dataset"

### Regression

* "Build a linear regression model predicting house prices"
* "What factors most influence customer satisfaction scores?"

### Classification

* "Train a model to classify emails as spam or not spam"
* "Predict which deals will close based on historical win/loss data"

### Dimensionality reduction

* "Reduce these 50 features to the most important 5"
* "Create a 2D visualization of these high-dimensional data points"

## How the AI uses ML

When you ask for a prediction or classification, the AI:

1. **Reads your data** from the spreadsheet
2. **Selects features** — determines which columns are inputs vs. targets
3. **Handles preprocessing** — missing values, encoding, scaling
4. **Trains the model** (or uses TabPFN for zero-shot prediction)
5. **Returns results** — predictions, confidence scores, feature importance
6. **Writes back** to your spreadsheet

## Available libraries

| Library          | Capabilities                                                     |
| ---------------- | ---------------------------------------------------------------- |
| **TabPFN**       | Zero-shot prediction and classification for tabular data         |
| **scikit-learn** | Classification, regression, clustering, dimensionality reduction |
| **SciPy**        | Statistical tests, distributions, optimization                   |
| **StatsModels**  | Regression analysis, time series, hypothesis testing             |
| **NumPy**        | Numerical computing, linear algebra                              |
| **pandas**       | Data manipulation and preprocessing                              |

## Use cases

| Scenario              | Approach                                        |
| --------------------- | ----------------------------------------------- |
| Sales forecasting     | Predict mode with historical sales data         |
| Churn prediction      | Classify mode with customer activity data       |
| Customer segmentation | scikit-learn clustering                         |
| Price optimization    | Regression with feature importance              |
| Anomaly detection     | Classify or cluster with outlier identification |
| A/B test analysis     | StatsModels hypothesis testing                  |
| Demand forecasting    | Time series with StatsModels                    |