User Guides

Choosing a Service

Match your data and goal to the right workflow.

Quickstart
Get from sign-in to a completed first job.
Choosing a Service
Match your data and goal to the right workflow.
Data Preparation
Prepare tabular datasets and media manifests for successful uploads.
Jobs and Results
Understand statuses, reports, SHAP, and outputs.
Billing and Credits
How charges, estimates, and refunds work.
Prediction Guide
Use trained jobs for manual and batch prediction.
Troubleshooting
Resolve common user-facing errors quickly.
FAQ
Short answers to the most common platform questions.
Reference
Handbook and Technical Reference

Open the handbook for the full product guide, then use the model-method appendix when you need mathematical details.

Open Handbook Open model-method appendix

Choosing a Service

Choose by two questions

The easiest way to choose a workflow is to answer: 1. What shape is my data? 2. What kind of outcome do I want?

Start with your data shape

  • Spreadsheet or cross-sectional data: supported tabular dataset rows where each row is one record
  • Longitudinal data: repeated observations for the same subject over time
  • Time series: ordered sequences where temporal order must be preserved
  • Text: documents, messages, transcripts, or other free text
  • Image, audio, video: media classification workflows
  • Graph or network: nodes, edges, or graph-derived features
  • User-item data: interactions used for recommendation or ranking

Then choose by analytical goal

  • Classification: predict categories such as yes/no, risk level, or label
  • Regression: predict a continuous number such as cost, score, or measurement
  • Count prediction: predict counts such as visits, events, or claims
  • Time-to-event: model survival or failure timing with censoring
  • Anomaly detection: flag unusual or rare records without a standard target
  • Clustering: group similar records without a standard target
  • Ranking or recommendation: order or suggest items by relevance

Quick mapping

  • Spreadsheet or cross-sectional data plus classes = Cross-Sectional Classification
  • Spreadsheet or cross-sectional data plus numeric outcome = Cross-Sectional Regression
  • Spreadsheet or cross-sectional data plus counts = Cross-Sectional Count Regression
  • Repeated measures plus classes = Longitudinal Classification
  • Repeated measures plus counts = Longitudinal Count Regression
  • Ordered sequences plus future or continuous values = Time Series Regression
  • Ordered sequences plus labels = Time Series Classification
  • No target plus unusual-record detection = Anomaly Detection
  • No target plus grouping = Clustering
  • Time-to-event with censoring = Survival Analysis

Important naming note

The current product uses Cross-Sectional in user-facing navigation for one-row-per-record workflows. You may still see tabular in some URLs, logs, or backend identifiers while the technical naming catches up.

Decision rules that help avoid wrong starts

Use classification when

  • the outcome is a label
  • you care about accuracy, recall, F1, ROC AUC, or PR AUC

Use regression when

  • the outcome is continuous
  • you care about RMSE, MAE, R^2, or forecasting error

Use count regression when

  • the outcome is a nonnegative count
  • zeros and event frequency matter
  • deviance or count-specific fit matters

Use anomaly detection when

  • you mostly need to flag unusual cases
  • you do not have a trusted supervised label for every record

Use clustering when

  • you want segments or groups
  • you do not have a standard target column

Use survival when

  • the question is about time until event
  • censoring matters

When to use the full Services page

  • you are comparing more than one possible workflow
  • you want to inspect required columns before upload
  • you need to confirm whether prediction, SHAP, or manuscript outputs are supported

Common confusion points

  • Anomaly detection is not the same as classification, even if anomaly labels exist later for evaluation
  • Clustering does not require a standard target column
  • Recommendation and ranking are not generic regression tasks
  • Time-series workflows are not just a cross-sectional spreadsheet with a date column
  • Survival analysis is not ordinary regression on time

If you are still unsure

Open the upload page for the two most likely services and compare: - required columns - example data format - readiness checks - expected outputs

The correct workflow is usually the one whose required columns and expected outputs match your intended analysis.

Related guides

  • Quickstart
  • Data Preparation
  • Jobs and Results
  • FAQ