Autots

List all cached objects across all cache types (predictions, autots_models, event_risk, feature_detectors, data). Call this to discover existing IDs before calling any get_*, plot_*, or apply_* tools.

No parameters. Call it with no arguments.

Clear cache: specific object by ID+type, an entire cache type, or all caches. Destructive — freed objects cannot be recovered.

Load a built-in sample time series dataset. Returns data_id for use as the data_id parameter in forecast_*, detect_features, forecast_event_risk, clean_data, and get_data.

Load live data from external sources (FRED economic data, stock tickers). Returns data_id. Requires network access.

Generate synthetic time series data with labeled components for testing. Returns data_id.

Load a CSV from a local file path or URL. Returns data_id. CSV must have a datetime column as the first column (index).

Retrieve cached data as JSON (wide or long) or save as CSV. Requires data_id from load_sample_data, load_live_data, load_data_from_file, generate_synthetic_data, convert_long_to_wide, or clean_data.

Convert long-format data (datetime, series_id, value columns) to wide format. Provide either data (inline dict) or data_id. Returns new data_id.

Clean time series data: fill missing values and handle outliers. Provide either data (inline dict) or data_id. Returns new data_id with cleaned data.

FAST: Pre-configured mosaic ensemble forecast using fit_data (no model search). Use for quick results. Provide data or data_id. Returns prediction_id for use in get_forecast, plot_forecast, apply_constraints, apply_adjustments, get_model_params.

MODERATE: AutoTS model search restricted to EXPLAINABLE models (Cassandra, TVVAR, BasicLinearModel). Use when interpretability matters. Provide data or data_id. Returns prediction_id and autots_id. Use get_forecast_components on prediction_id for component decomposition.

CUSTOM: AutoTS with user-specified parameters or template. Use when forecast_fast results are insufficient. Read the autots://docs/forecast_custom_params resource for available parameters. Defaults to 'scalable' model_list. Provide data or data_id. Returns prediction_id and au...

Retrieve forecast values from a cached prediction as JSON or CSV. Requires prediction_id from forecast_fast, forecast_explainable, forecast_custom, or forecast_from_features. Use output='all' to get point, upper, and lower forecasts combined with a forecast_type column.

Plot forecast with optional history and prediction intervals. Requires prediction_id from forecast_fast, forecast_explainable, forecast_custom, or forecast_from_features. Returns base64-encoded PNG image. Defaults to first series only.

Apply constraints to a forecast (dampen growth, enforce upper/lower bounds, or quantile clipping). Requires prediction_id from a forecast tool. Returns a new prediction_id with constrained values.

Apply post-hoc adjustments to a forecast. Requires prediction_id from a forecast tool. Three types: 1) 'basic'/'linear'/'ramp' — linear ramp between start/end dates with start/end values (additive or multiplicative); 2) 'align_last_value'/'alignlastvalue' — align forecast to r...

Get model name, parameters, and transformation parameters from a cached prediction. Requires prediction_id from any forecast tool.

Get decomposed forecast components (trend, seasonality, etc.) if available. Only works for Cassandra and TVVAR models — use forecast_explainable to guarantee these model types. Requires prediction_id.

Get cross-validation results and top model rankings from an AutoTS model search. Requires autots_id from forecast_explainable or forecast_custom.

Plot cross-validation forecasts from an AutoTS model search. Requires autots_id from forecast_explainable or forecast_custom. Returns base64-encoded PNG.

Forecast the probability of crossing a threshold over future periods (e.g., stockout risk, capacity breach). Threshold in [0,1] is treated as a historical quantile; values outside [0,1] are absolute. Provide data or data_id. Returns event_risk_id for use in get_event_risk_resu...

Get event risk probability values from a cached EventRiskForecast. Requires event_risk_id from forecast_event_risk. Returns probabilities per period per series.

Plot event risk probabilities over the forecast horizon. Requires event_risk_id from forecast_event_risk. Returns base64-encoded PNG.

Detect anomalies, changepoints, level shifts, holidays, and seasonality patterns across all series. Provide data or data_id. Returns detector_id for use in get_detected_features, plot_features, and forecast_from_features.

Query detected features (anomalies, changepoints, level shifts, holidays, seasonality) from a cached detector. Requires detector_id from detect_features. Supports filtering by date range, specific date, or series name. Use include_components=true for time-series component valu...

Plot detected features overlaid on the time series. Requires detector_id from detect_features. Returns base64-encoded PNG.

EXPERIMENTAL: Create a forecast using the decomposed components from a feature detector. Only use after detect_features, not as a standalone forecasting method. Requires detector_id from detect_features. Returns prediction_id for use in get_forecast, plot_forecast.