Importing and editing data

Equarith works with immutable numerical dataset snapshots. You can import a delimited text file, paste tabular data, edit it in the virtual grid, or load a generated sample. Search and statistics use a stable snapshot, so background work never reads a half-edited grid.

Ways to load data

• Select Import File in the Input workspace, use File > Import Dataset…, or press the import shortcut to choose a CSV, TSV, or numerical text file.
• Drag and drop a .csv, .tsv, or .txt file onto the Input workspace to open the same import workflow.
• Select Paste Data to read a tab-separated table from the clipboard. With an existing editable dataset, the values are inserted from the current grid selection; without one, the clipboard becomes a new dataset.
• Select Load Example to generate a small demonstration dataset locally. At normal startup, Braking Distance is already loaded.

Importing or pasting replaces or modifies the project data only after validation. It never modifies the source file or clipboard content.

Supported text data

The importer accepts CSV, TSV, and other delimited numerical text files. It can detect:

• comma, semicolon, tab, or whitespace field separation;
• whether the first row is a header;
• dot or comma decimal notation; and
• common text encodings.

The import dialog shows Data preview and Preview diagnostics. Review or override Delimiter, Header row, Decimal style, Encoding, and Quote character, then select Import. Common encoding choices include UTF-8, ISO-8859-1, and US-ASCII; any charset supported by the local Java runtime can be entered.

The preview displays at most five individual diagnostics and summarizes any additional sampled issues. It is intentionally bounded for responsiveness. The complete file is parsed and validated only when you select Import, before the active dataset is replaced.

A comma cannot be both the field delimiter and decimal separator. For decimal-comma data, use semicolon, tab, or whitespace separation.

Missing, malformed, and non-finite values

By default, a blank field, NA, or NaN is treated as missing without regard to letter case. Leading and trailing whitespace is ignored and completely blank rows are skipped.

Positive infinity, negative infinity, and other non-finite numeric values are invalid for search. Rows with too few fields are padded and rows with too many fields are truncated, with diagnostics shown to the user. An invalid quoted structure is treated as a file error because field boundaries can no longer be interpreted safely.

Review the final row, column, missing-value, and diagnostic counts rather than relying only on the preview.

Headers and formula symbols

Column display names are converted into safe, unique symbols when formulas are built. Prefer short, distinct names such as mass, velocity, and distance. Put units in documentation or use concise suffixes when needed, for example time_s.

Renaming a column updates the data model while preserving its stable identity where the format requires it. Equarith prevents ambiguous formula symbols even when imported headers repeat or contain punctuation.

Editing the grid

The data grid is virtualized and remains practical for datasets that would be too large to represent with one UI control per cell.

• Double-click a cell, or press Enter or F2, to edit it.
• Use arrow keys to move.
• Shift-select ranges and use the platform shortcut with C or V to copy or paste.
• Pasted table data is tab-separated.
• Add rows or columns from the grid controls.
• Rename a column from its header.
• Press Delete or Backspace to clear selected editable cells. Delete complete rows or columns with the appropriate command or context menu.
• Undo and redo edits with the normal platform shortcuts.

Changes are applied outside the JavaFX application thread. While a search is active, importing, pasting, cell and structure editing, undo, and redo are disabled so its captured snapshot and the visible project cannot diverge. Stop the search, edit the data, then select Start again to use the new snapshot.

Which rows participate in search

A row is eligible only when the target and every selected input have finite numeric values. Equarith excludes other rows before making the train/test split.

In Demo mode, the row limit is applied to the first 200 source rows before eligibility filtering. This means 200 source rows can produce fewer than 200 usable search rows. Demo mode also accepts no more than four selected inputs for a search.

History functions require preceding source rows. Their warm-up rows and rows with invalid required history values are excluded before splitting. Random train/test assignment does not change the chronological order used to calculate history.

The general dataset model requires at least three rows and two columns, but a particular search also needs enough eligible rows for its requested split. Practical maximum size depends on available memory, column count, enabled operations, and worker count. Equarith also applies hard safety bounds to reject unreasonable dense datasets.

Data preparation guidance

• Inspect missing and invalid counts before selecting variables.
• Keep units consistent within each column.
• Avoid identifiers, dates, and categories encoded as arbitrary numbers unless their numerical meaning is deliberate.
• Look for accidental duplicates, impossible values, and changes of measurement regime.
• Use a sequential split for genuinely time-ordered validation; use a seeded random split for interchangeable observations.
• Do not remove outliers automatically. Investigate whether they are errors, rare valid cases, or evidence that one formula is inadequate.
• Normalization can improve numerical search while published formulas are converted back to the original units.

Persistence and export

An unchanged imported file can be stored in a project by reference, or its data can be embedded for portability. Edited, pasted, and generated data is embedded. Referenced data is checked by content fingerprint and schema when the project is reopened.

Select Export CSV to write the current edited dataset. Clear Dataset removes the entire dataset from the project after confirmation; it does not delete an imported source file. Formula prediction exports are separate: Prediction-workspace CSVs include one status column per formula, while Search and CLI prediction CSVs contain the source columns and prediction, using an empty prediction cell for a non-finite result.

Continue with Configuring a search or read Projects, recovery, and checkpoints.