How to improve predictions

Training your Collection

When you start extracting data from your file, Impira immediately starts looking for matching data in the rest of the files in your Collection. Impira will then display values that appear to match your added fields — these are called predictions. Predictions can vary in confidence level and accuracy at first.

You can then "train" your Collection to increase the confidence level of predictions and improve prediction accuracy.

You can train a Collection by:

  • Confirming predictions
  • Correcting predictions
  • Adding more examples of a data field on more files

Continue reading to learn how to improve predictions of various types: Text, table, and checkbox.

Text predictions

Confirming text predictions

  1. Open a file in Extraction view and hover over a prediction in the right sidebar.
  2. Click the checkbox to confirm a correct prediction.

Editing text prediction values or modifying the blue bounding box

When in Extraction view, click on a prediction and edit the value in two ways:

  1. Best method: Adjust the corners of the blue box so it correctly fits over the value. Adjusting the blue box will automatically change the value that shows up in the value box and will retrain Impira.
  2. Or, edit the text itself by typing in the correct value. (Fun fact: The blue box will automatically readjust its shape to the text you've typed.)

Click Confirm value and blue box after you're done making edits.

Table extraction predictions

With table extraction, it's important to confirm entire rows of predictions. This will signal to Impira to reprocess predictions based on new learnings.

  1. Open the row in Extraction view in the right sidebar.
  2. Correct any text values in the row as necessary.
  3. Hit Confirm entire row or confirm each value in the row.

Troubleshooting tip: If your table extraction results are showing a lot of unintended rows, check to see if you've selected the correct value type (Text, Number, or Date).

The Text type accepts anything (like letters, numbers, and symbols). This is the most flexible type but doesn't give Impira a lot of specific information.

The Number type will only accept numbers (and related symbols like “-” and currency). Choosing this will tell Impira to ignore blocks of text for these values.

The Date type only accepts dates that contain a Day, Month, and Year in formats that are less likely to be misinterpreted, like “Jan. 1, 2022” or “2022-01-01” (if this last type causes errors, you'll have to manually enter the date in a less ambiguous way).

Value types can't be changed after the column is created. If you have the wrong type, delete the column with the incorrect type and create a new one with the correct type.

Checkbox predictions

Improve your checkbox predictions by:

  • Confirming correct ones (both high and low confidence)
  • Correcting incorrect predictions
  • Inputting values for “No value detected” predictions.

Confirming correct checkbox predictions

In Extraction view, hover over a correct prediction and click the Confirm predicted value checkmark.

Correcting incorrect or low confidence predictions

  1. In Extraction view, click the checkbox value in the right sidebar.
  2. If the blue box is shaped incorrectly, readjust the corners so it fits properly over the checkbox.
  3. Go to the right sidebar, ensure that the value is correct (checked or unchecked), and select Confirm value and blue box when you're done.

Inputting values for “No value detected” predictions

  1. Click the “No value detected” prediction.
  2. Make any necessary changes to the checkbox value and/or the blue box.
  3. Click Confirm value and blue box when you're finished.

© 2022 Impira Inc. All rights reserved. This site is built with Motif.