Formula Rules and Transform Instructions

Formula Rules

How to Create a Custom Formula for Synthetic Data:

1. Referencing Columns: Use column_# to reference columns (e.g., column_1, column_2).
2. Basic Operations: Supported operators: +, -, *, /, ^ (for power).
3. Math Functions: Use log(), exp(), sqrt(), abs(), sin(), cos(), tan(), etc.
4. Random Numbers:
   • random(): Generate numbers between 0 and 1.
   • random(min, max): Generate numbers within a range (e.g., random(0, 10)).
5. Cumulative Sum:
   • cumsum(column_#): Compute the cumulative sum of the values in the specified column.
6. Differences:
   • diff(column_#): Compute the differences between consecutive values in the specified column, with the first value set to 0.
7. Conditional Logic: Use if(condition, value_if_true, value_if_false).
   • Example: if(column_1 >= 2000, 1, 0)
   • Example: if(column_1 == nan, 1, 0)
8. Constants: Include numeric values directly (e.g., 42, 3.14).
9. NaN Values: Undefined values or empty cells are replaced with -0.9999.
10. Dynamic Column Naming: When generating synthetic data, a new column is created using the provided property name. If the provided property name already exist in the table as a column, it will replace that column.

Examples:

• column_1 + 5
• column_1^2 + column_2 * 3
• if(column_1 > 1000, 0, 1)
• random(0, 100)
• 3.1415926

Transform Instructions

• Remove Missing Data: Remove rows with missing value in selected columns. Empty cells or placeholder values (-0.9999) are treated as missing value.
• Generate Synthetic Data: Create a column with the name you select or enter using predefined or custom formulas according to the rules above.
• Transform with Neural Networks:
  • Training: Train a neural network model using selected features and target. The trained neural network model will be saved and be used to make prediction.
  • Predicting: If only features are selected and no target is selected, the last trained neural network model will be used to make a prediction. The prediction result will be saved into the table as "Predicted_Target".