To ensure accurate predictions using an already trained model, follow these steps to create a compatible dataset with metrics:

1. Match the Training Dataset Structure

Your prediction dataset must exactly replicate the structure of the dataset used during model training. This includes:

• Metrics:Ensure all metrics are calculated using the same formulas as in the training dataset.
• Columns:Maintain the same column names, formats, and sequence.
• Data Scope:Include only data for the last available observation date (date_observation).

2. If the Model Was Trained Using a .csv File

• Create a .csv file with metrics that mirror the training dataset in every aspect.
• Double-check the consistency of column names, data types, and calculated values.

3. If the Model Was Trained Using a Google BigQuery (BQ) Table

You have two options for creating the prediction dataset:

• Option 1: Extend the Original Table
  • Add rows with new data (users and their metrics) to the table used during training.
  • The system will automatically select rows with the latest date_observation for predictions.
• Option 2: Create a New Prediction Table
  • Set up a new table in BigQuery that mirrors the structure of the training dataset:
    • Use identical metrics, column names, and formats.
  • Configure a regular update schedule for this table:
    • Calculate metrics periodically based on all available data up to the analysis date.
    • This ensures predictions are based on the most up-to-date information.

4. Switching from .csv to BigQuery (Optional)

If you initially trained the model using a .csv file but now want to use BigQuery for predictions: