Skip to main content
You don’t need to worry about aligning your inference data with your training data. Wood Wide AI automatically learns the schema (column names and data types) from your training dataset and reconciles any differences at inference time. This means you can send inference data that is slightly different from what you trained on, and it will just work:
  • Extra columns are ignored. Include additional columns in your inference data and they won’t affect results.
  • Missing columns are fine. If your inference data is missing a column that was in the training data, it’s filled in automatically.
  • Type differences are resolved. If a column was numeric during training but arrives as strings during inference, non-numeric values are converted gracefully.
  • New categorical values are handled. If a categorical column sees values at inference time that weren’t present during training, the model handles them without errors.
In all cases where a value can’t be reconciled, it is treated as missing data internally. The model is designed to handle missing values, so this won’t cause failures.

​
Technical Details

At training time, Wood Wide AI infers the schema of your dataset — the column names and their broad data types (numeric, categorical, datetime, binary). The trained model is tied to that schema. When you run inference, the input data is automatically coerced to match the training schema:
  • Extra columns not in the training schema are silently dropped.
  • Missing columns are filled with null values.
  • Type mismatches are coerced where possible. For example, if column A was numeric at training time but contains strings at inference time, non-numeric values are replaced with nulls.
  • Unseen categorical values are handled gracefully. If column B contained only values 0 and 1 at training time but the inference data also includes 2, 3, and 4, those unseen values are replaced with nulls internally.