Uploading spectra

Click Upload Spectra on the Spectra page to open the upload dialog.

Upload spectra dialog showing sensor picker and CSV file drop zone

The upload flow has up to four steps. Property extraction (step 2) is optional.

Spectra with the same sample name are treated as replicas, not duplicates.Unlike samples (where the same name updates the existing record), uploading multiple spectra with the same sample name creates multiple spectra all linked to the same sample. This is intentional: real lab workflows produce multiple measurements per sample (different runs, different days, different operators).If you upload 3 files all named Sample-001.csv, you get one sample with three spectra attached. All three are used together in datasets and models.

Step 1: select sensor and files

Choose the sensor that produced the files. The dropdown shows all sensors registered in this project, formatted as name — manufacturer model. Drag and drop your CSV files into the upload zone, or click to browse. Multiple files at once are supported.

If your sensor was created from the catalog, Chemolytic uses the built-in importer for that instrument’s raw file format.If your sensor is custom, your files must follow the Chemolytic standard CSV format (see below).

File requirements

Requirement	Value
File extension	`.csv`
Encoding	UTF-8 preferred. Chemolytic also accepts UTF-8 with BOM and Latin-1 as fallback.
Files per upload	Multiple files allowed in one batch

Catalog sensor parsers

If your sensor was created from the catalog, just upload the raw files exported by your instrument software. Chemolytic detects the format from the sensor model and reads the file directly, no reformatting needed. The filename (without .csv) becomes the sample name.

Manufacturer / model	Sensor models supported	File layout
InnoSpectra NIRscan	`NIR-M-R2`, `NIR-S-G1`, `NIR-M-R3`, `NIR-M-T1`, `NIR-M-13`, `DLP NIRscan`	One spectrum per file. Metadata header (key-value pairs) followed by a column header and two data columns (x, y).
Si-Ware NeoSpectra	`NIR-M-R21`	One spectrum per file. Si-Ware native CSV export.
PerkinElmer	`Spectrum 3`	One spectrum per file. PerkinElmer Spectrum software CSV export.

If your sensor model is not in this table but your instrument exports a CSV that matches one of the standard layouts below, create a custom sensor and upload using one of those formats.

Standard CSV format (custom sensors)

Custom sensors accept two layouts. Chemolytic auto-detects which one your file uses.

Long format (one row per data point)

Three columns by position. The first row may be a header (skipped automatically if not numeric).

Column	Notes
`sample_name`	The name of the sample this row’s data point belongs to
`x`	Wavelength or wavenumber value
`y`	Intensity value at that wavelength

A single file can contain multiple samples. Each sample is identified by its sample_name. One row per data point per sample.

sample_name,x,y
Sample-001,950.0,0.245
Sample-001,952.0,0.251
Sample-001,954.0,0.258
Sample-002,950.0,0.301
Sample-002,952.0,0.298

Wide format (one row per sample)

Header row starts with the literal word name (case-insensitive) followed by the x-values. Each subsequent row is one sample: name in column 1, y-values in the remaining columns aligned with the header x-values.

name,899.84,900.45,901.06,901.67
Sample-001,0.4231,0.4258,0.4287,0.4310
Sample-002,0.3987,0.4012,0.4045,0.4071

Sample names must be unique within the file. Every data row must have the same number of y-values as the header has x-values.

Both formats can be mixed across uploads to the same sensor, as long as the x-axis matches the sensor’s locked template (see Sensor x-axis lock).

Click Next to proceed.

Step 2: property extraction (optional)

If your filenames contain useful information about the samples (e.g., oliveoil-batch12-arabica.csv), you can extract that information into sample properties automatically.

Property extraction step showing delimiter picker and a position assignment table

How extraction works

Pick a delimiter that separates the parts of your filenames: dash (-), underscore (_), dot (.), comma, space, or a custom character
Chemolytic splits each filename by the delimiter, showing every position in a preview table
For each position, choose Skip or Extract
For positions you extract, pick (or create) the property to fill, and set the type (Continuous or Categorical)

Example

For files named oliveoil-batch12-arabica.csv and oliveoil-batch13-robusta.csv, with delimiter -:

Position	Sample preview	Action	Property	Type
1	oliveoil	Skip
2	batch12, batch13	Extract	Batch	Categorical
3	arabica, robusta	Extract	Variety	Categorical

This creates two new properties (Batch and Variety) and assigns the right value to each sample automatically.

Properties created during extraction follow the same rules as manually created properties. New property names must be alphanumeric (with underscores allowed). Categorical properties need at least 2 unique values across your files.

Errors at this step

Error	Cause
`Name at least one position to extract a property.`	You toggled extraction on but didn’t name any property
`Duplicate property name: "{name}"`	Two positions are extracting into the same property

Skip extraction

If your filenames don’t contain useful information, click Skip. Each file’s name (without .csv) becomes the sample name and that’s it. Click Parse & extract (or Skip) to proceed.

Step 3: review and select

The review step has two tabs.

Properties tab

Shows the extracted property values for each sample. You can:

Edit any cell to fix incorrect extractions
See warnings for samples where extraction failed (highlighted amber)
Use the search box to find specific samples
Select or deselect samples to import (checkbox per row)

Review tab showing extracted property values per sample with editable cells

Spectra tab

Shows the spectra chart with all parsed spectra overlaid, plus a chip per spectrum below.

Review tab showing the spectra chart and selectable chips for each spectrum

You can:

Click a chip to deselect a spectrum (it shows strikethrough and is removed from the chart)
Use Select all or Deselect all to bulk toggle
Search by sample name

Click Import N spectra to confirm.

Errors at this step

Error	Cause
`X axis has N points but sensor expects M.`	The file’s x-axis has a different number of points than the sensor’s locked template
`X values differ from sensor template at K point(s)`	The x-axis values don’t match the sensor’s locked template within tolerance
`'{sample}' has N x points but '{first}' has M.`	Files in the same batch must share the same x-axis
`Property '{prop}' already exists as 'categorical' but extraction infers 'continuous'.`	Existing property type doesn’t match what extraction would create
`Property '{prop}' is continuous but value '{val}' for sample '{sample}' is not numeric.`	Extracted value cannot be parsed as a number for a continuous property

Step 4: success

After import, you see a summary of everything that was created and linked.

Import complete screen showing summary counts for spectra, samples, matched, properties, and targets

Summary fields

Spectra created The total number of spectra saved to the project from this batch. This counts every file you imported, including replicas (multiple files for the same sample). Each spectrum is independently archivable and deletable. New samples Samples that did not exist before this upload and were auto-created. Chemolytic uses the filename (or the sample_name column for the standard CSV format) to determine the sample name. If a name doesn’t match an existing sample, a new one is created with no description and no property values (unless extracted at step 2). Matched Spectra whose sample name matched an existing sample in the project. The new spectra are linked to that sample as additional replicas. The existing sample’s description and other property values are not modified. Properties New sample properties created during the extraction step. Only appears if you used property extraction at step 2 and at least one new property was created. Existing properties are reused, not duplicated. Targets The total number of property-value pairs written to samples. One target = one (sample, property, value) triple. If a sample already had a value for that property, it is updated; otherwise a new target is created.

What happens next

Click Done to close the dialog and return to the Spectra page. The new spectra appear in the table immediately. You can now:

View any spectrum by clicking its row
Filter by sensor to see the new spectra in the chart preview
Continue to Building a dataset to use this data for training

Sensor x-axis lock

The first spectrum uploaded for a sensor sets that sensor’s x-axis (number of points and range). All subsequent uploads must match. If you upload a file with a different x-axis to an existing sensor, the parse step fails with one of the errors listed above. You have two options:

Fix the file (resample to match the sensor’s x-axis), or
Create a new sensor for the different x-axis configuration

The x-axis lock cannot be unset. If you mistakenly upload a file with the wrong x-axis as the first spectrum, you must create a new sensor and start over.

Plan limits

The plan limit on spectra per project is checked at the confirm step. If your batch would exceed the limit, the import is rejected before any spectra are saved.

Tips

Test with one file first. Upload one file to confirm the parser works for your instrument before importing a large batch.

Use consistent filenames. If you plan to extract properties from filenames, keep the structure consistent across all files (same delimiter, same number of positions, same order).

Re-uploading is safe for samples. If a sample already exists with the same name, the new spectra are linked to it (the sample is not duplicated).

Getting started

Account & management

Hardware

Data

Exploration

Modelling

Production