Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.chemolytic.com/llms.txt

Use this file to discover all available pages before exploring further.

The web prediction interface is the simplest way to use a deployed model. Upload a CSV of new spectra, get predictions back in a table. Open any active deployment and click the Predict tab.
Predict tab showing sensor selector, file upload zone, and a results table with sample names and predicted values

​
Step 1: pick a sensor

The sensor dropdown shows all sensors in the project. Each option includes the sensor name, model name, and number of points.
The sensor must match the one the model was trained on. Specifically, the point count and x-axis values must match. Predictions on spectra from a different sensor will fail with a clear error message.

​
Step 2: upload spectra

Drag and drop CSV files into the upload zone, or click to browse. Multiple files at once are supported. The format depends on the sensor:
  • Catalog sensor: use the raw file format your instrument exports
  • Custom sensor: use the standard CSV format (long format with sample_name, x, y columns)

​
Step 3: run predictions

Click Run predictions. The button is disabled until you have both a sensor and at least one file selected. While predictions are running, the button shows a loading state. Behind the scenes, Chemolytic:
  1. Parses each file using the sensor’s importer
  2. Validates the point count matches the model’s expected input
  3. Sends the spectra to the model
  4. Returns one prediction per spectrum

​
Step 4: read the results

A results table appears below the upload zone.
ColumnDescription
SampleSample name extracted from the file
Predicted valueThe model’s prediction (number for regression, category for classification)
For regression models, values are shown to 4 decimal places. For classification, the predicted category is shown as a label.

​
Downloads

Two buttons appear above the table:
ButtonFormat
CSVTwo columns: sample_name, predicted_value
JSONObject with a predictions array
Use these to bring predictions into your downstream workflow (LIMS, Excel, custom report).

​
Validation errors

The CSV doesn’t match the model’s expected input size. Either you picked the wrong sensor, or the file is from a different instrument configuration. Use the sensor that matches the file.
The deployment is currently inactive. Toggle it active in the Overview tab and try again.
The underlying model is still building or failed. Wait for it to finish (or check the model detail page for an error).
You’ve used all your web predictions for this month. The error message shows your limit, used count, and how much you tried to submit. Wait for the next billing cycle or upgrade your plan.
One of your spectra contains invalid numbers. Check the source file; this is usually a parsing or instrument issue.

​
Plan limits

Each web prediction counts against max_webapp_predictions_month. The current count and limit are shown at the top of the deployment detail page. When you exceed the limit mid-batch, the entire batch is rejected (you don’t get partial results). The error message shows exactly how many you tried to submit and how many remain.
For high-volume predictions, use the API instead. It has its own monthly quota and is faster for bulk requests.