Prepare your data
Structure your dataset for unsupervised, supervised, longitudinal, or external-result analysis before uploading it to Ounias.
Basic dataset structure
Ounias works with structured tabular data: rows are observations and columns are variables.
Datasets must contain at least 8 observations and at least 3 feature columns.
Supported file formats are CSV, XLS, and XLSX. Files can be uploaded locally or from Google Drive.
For Excel workbooks containing multiple sheets, choose the sheet to import after upload. Each imported sheet is treated as a separate dataset.
- One row = one observation
- One column = one variable or feature
- At least 8 rows
- At least 3 feature columns
- Use clear, unique column names
Use stable identifiers
Stable anonymized identifiers make observations easier to track, reproduce, and match with external results.
For standard datasets, provide a unique anonymized row ID when possible. Ounias can generate observation IDs automatically when none are provided.
If you expect to upload external model results later, providing your own stable IDs is strongly recommended. Ounias-generated IDs can be exported, but matching them back to another system later may be inconvenient.
Identifiers should not contain unnecessary personally identifying information.
Unsupervised datasets
Each row represents an independent observation described by at least three analytical features.
A unique row ID is recommended but not required. If no identifier is provided, Ounias can generate one automatically.
Features may include continuous, categorical, or binary variables.
Supervised datasets
Supervised analysis uses the standard dataset structure together with a target column.
Regression targets should contain numeric values.
Classification targets contain class labels and may be represented using text or numeric values.
Ordered variables can be supplied using numeric values and treated as continuous features when preserving their ordering is appropriate.
Longitudinal datasets
Longitudinal datasets contain multiple observations for the same entity over time or ordered occurrences.
Provide an anonymized entity ID that repeats across observations belonging to the same subject or entity.
Each observation must also contain either an observation index or an observation datetime so individual rows can be distinguished.
Datetimes are preferred when available because they allow Ounias to account for the actual time intervals between observations.
Rows do not need to be manually sorted before upload; the selected ordering column defines longitudinal order.
- Entity ID identifies the subject or entity
- Observation index or datetime identifies the occurrence
- Entity ID + observation order must uniquely identify each longitudinal observation
Dates and datetimes
Use consistent date or datetime values when temporal spacing matters.
Ounias parses recognizable date and datetime values automatically.
Consistent ISO-style values such as 2026-09-26 or 2026-09-26T14:30:00Z are recommended to reduce ambiguity.
External model results
External results are uploaded separately and matched back to observations in an existing Ounias dataset.
For a standard dataset, provide exactly two columns: the original row ID and the external result.
For a longitudinal dataset, provide exactly three columns: entity ID, the original observation index or datetime, and the external result.
Every observation in the target Ounias dataset must have a corresponding external result. Extra uploaded observations may be dropped automatically.
File order does not matter. Ounias matches observations using their identifiers and restores canonical dataset order internally.
- Classification: use the same class labels as the original target
- Regression: results must be finite numeric values
- Clustering: numeric or text cluster labels are accepted; non-integer labels are normalized automatically
- Identifiers, ordering values, and results cannot be missing
- Duplicate observation identifiers are not allowed
What Ounias handles for you
You generally do not need to prepare separate model-specific versions of your dataset.
After import, Ounias validates feature roles and prepares analytical representations required by downstream methods.
- Missing-feature handling
- Cleaning and normalization where required
- Categorical and binary feature handling
- Method-compatible transformations
- Automatic observation IDs when none are supplied