Prediction

This page describes Step 6 in the dAIve web app.

Left panel

The prediction controls are grouped into:

• Your trained model with a Choose .dvm file drop zone
• Import Prediction Data with a Choose .csv file drop zone
• file separator and decimal separator settings

If the project already contains a trained model, the page can use it directly. The model can also be replaced by importing another .dvm file via drag-and-drop or file browser.

What must stay consistent with training:

• input column names
• input order where applicable
• decimal and file separator interpretation
• time-series expectations for RNN models

Running predictions

The main action button changes label depending on the state:

• Run Prediction when no predictions exist yet
• Update Prediction when the table already contains results
• Stop Prediction while a prediction is running

The button is disabled when no prediction data file has been loaded and no manual row values have been entered.

Batch prediction

Use a CSV when you want predictions for many rows at once.

Workflow:

1. load the model if needed
2. upload the prediction CSV via the drop zone
3. adjust separators if needed
4. click Run Prediction or Update Prediction
5. review the prediction table
6. use Export table to download results as a CSV file

Use batch prediction when:

• you have many records to score at once
• you need a downloadable output file
• you want the same model applied to a prepared dataset

Manual what-if analysis

The prediction table also supports manual input rows (highlighted with a blue tint).

For each input column:

• numeric columns show a text input with a "Manual value" placeholder
• categorical columns show a dropdown selector with a "Select value" placeholder and the known class options from training

Output columns in manual rows display the predicted values after running the prediction.

Use manual rows to:

• enter values without uploading a CSV
• compare several scenarios
• test what happens when inputs change

Manual rows are useful when:

• you need quick scenario testing
• you want to understand model sensitivity
• you are validating a few candidate cases before a larger batch run

RNN-specific option

If the current model is an RNN, the page also shows:

• Future Steps to Predict (number input, 1–100, default: 1)

This controls how many future time steps the model predicts ahead.

Table warnings

The prediction table can mark values outside the learned value range with a red background. These rows are extrapolation candidates and should be interpreted carefully.

Treat those warnings as a sign that:

• the input may be outside the distribution seen during training
• the prediction may be less reliable than in-range cases

Invalidation

When earlier steps change (e.g. inputs, outputs, or model configuration), a yellow warning badge appears on the prediction step. This means the current predictions may be stale and should be updated.

Web-specific note

Prediction in the web app consumes compute credits (0.25 credits per prediction run).