datagrok-df-and-columns

Use the DG.DataFrame / DG.Column js-api via the datagrok_exec tool. Globals available in the code: grok, ui, DG, view, t (the current DG.DataFrame, when view.type === 'TableView').

For formula-driven columns (recompute on source change), use datagrok-calc-column. Row filtering / selection lives in datagrok-filtering and datagrok-selection.

Quick reference

Need	Call
Get column by name (must exist)	t.getCol(name) — throws on missing
Get column by name (may be absent)	t.col(name) — returns null
First column with a given semType	t.columns.bySemType(DG.SEMTYPE.MOLECULE)
Every column with a given semType	t.columns.bySemTypeAll(DG.SEMTYPE.MOLECULE)
All numeric / categorical / dateTime columns	t.columns.numerical / .categorical / .dateTime
All names	t.columns.names()
Add typed empty column	t.columns.addNewFloat(name) (and addNewInt/String/Bool/...)
Add column from values	t.columns.add(DG.Column.fromList(type, name, values))
Virtual column (compute-on-demand)	t.columns.addNewVirtual(name, i => ..., type)
Collision-free name	t.columns.getUnusedName(name)
Remove a column	t.columns.remove(name) (string, index, or Column)
Rename	col.name = 'new'
Clone DataFrame (full)	t.clone()
Clone filtered rows	t.clone(t.filter)
Clone column subset	t.clone(null, ['a', 'b'])
Check if value is missing	col.isNone(i)
Stats (cached on column)	col.stats.{min,max,avg,stdev,med,q1,q2,q3,sum,valueCount,missingValueCount,uniqueCount}
Set semType	col.semType = DG.SEMTYPE.MOLECULE
Set friendly name / units / format / desc	col.meta.friendlyName / .units / .format / .description = ...
Linear color coding (numeric)	col.meta.colors.setLinear() (default scheme) or setLinear(range, opts?)
Categorical color coding	col.meta.colors.setCategorical() (default palette) or setCategorical(map, opts?)
Conditional color coding	col.meta.colors.setConditional() (auto bins) or setConditional(rules)
Disable color coding	col.meta.colors.setDisabled()

Finding the right column

User said...	Right call
"column named X" (must exist)	t.getCol('X') — throws if missing
"column named X" (may be absent)	t.col('X') — returns null if missing
"the molecule column"	t.columns.bySemType(DG.SEMTYPE.MOLECULE)
"every molecule column"	t.columns.bySemTypeAll(DG.SEMTYPE.MOLECULE)
"all numeric columns"	iterate t.columns.numerical
"an IC50 column"	t.columns.bySemType(DG.SEMTYPE.IC50)
"columns tagged quality=Molecule"	t.columns.byTags({quality: 'Molecule'})
"any of these columns: X, Y, Z (first hit)"	t.columns.firstWhere(c => names.includes(c.name))
"does the DF have a column named X?"	t.columns.contains('X') (case-insensitive)

t.col and t.columns.byName are case-insensitive. Always pass the semType constant (DG.SEMTYPE.MOLECULE), never the string literal 'Molecule'.

// Find the molecule column. bySemType returns the first match or null.
const molCol = t.columns.bySemType(DG.SEMTYPE.MOLECULE);
return ui.divText(molCol ? `Molecule column: ${molCol.name}` : 'No molecule column');

// User said "mw" but the DF has "Molecular Weight" — try exact first, then
// fall back to a name substring match.
const wanted = 'mw';
let col = t.col(wanted);
if (!col) {
  const lc = wanted.toLowerCase();
  col = t.columns.firstWhere((c) => c.name.toLowerCase().includes(lc));
}
return ui.divText(col ? col.name : 'no match');

Describing a column

col.stats is cached per column — repeated access is cheap.

Need	Use
total row count	col.length
non-null value count	col.stats.valueCount
missing-value count	col.stats.missingValueCount
distinct value count	col.stats.uniqueCount
min / max / avg / stdev	col.stats.{min,max,avg,stdev}
median / quartiles	col.stats.{med,q1,q2,q3}
sum	col.stats.sum
pairwise correlation	col.stats.corr(other)
sorted distinct strings	col.categories
typed-array raw view	col.getRawData()
value iterator (generator)	col.values()

Do not compute these by hand:

• Total rows: col.length, not col.toList().length (allocates).
• Non-null count: col.stats.valueCount.
• Distinct count: col.stats.uniqueCount, not new Set(col.toList()).size.
• For string columns, col.categories.length === col.stats.uniqueCount.

// Build a key-value summary of one numeric column.
const col = t.getCol('age');
const s = col.stats;
return ui.tableFromMap({
  name: col.name,
  type: col.type,
  semType: col.semType ?? '(none)',
  length: col.length,
  missing: s.missingValueCount,
  unique: s.uniqueCount,
  min: s.min, max: s.max, avg: s.avg, stdev: s.stdev,
});

// Summarize every numeric column. t.columns.numerical is Iterable<Column>
// (not an Array) — convert via Array.from to get .map etc.
const rows = Array.from(t.columns.numerical, (c) => ({
  name: c.name, min: c.stats.min, max: c.stats.max, avg: c.stats.avg,
}));
return DG.Viewer.grid(DG.DataFrame.fromObjects(rows)).root;

Top-N categories

No built-in top-N for a string column. Tally into a Map<string, number>:

const col = t.getCol('class');
const counts = new Map();
for (let i = 0; i < col.length; i++) {
  const v = col.get(i);
  if (v === null || v === undefined) continue;
  counts.set(v, (counts.get(v) ?? 0) + 1);
}
const top = Array.from(counts.entries())
  .sort((a, b) => b[1] - a[1])
  .slice(0, 5);
return ui.tableFromMap(Object.fromEntries(top));

If you only need distinct values (no counts), col.categories is already sorted alphabetically.

Cloning DataFrames

t.copy() does not exist. Use t.clone(rowMask?, columnIds?, saveSelection?, saveTags?):

Need	Call
Full copy	t.clone()
Filtered rows only	t.clone(t.filter)
Selected rows only	t.clone(t.selection)
Column subset (all rows)	t.clone(null, ['a', 'b'])
Filtered rows + column subset	t.clone(t.filter, ['a', 'b'])
Preserve selection bits	t.clone(t.filter, null, true)

// Just SMILES and activity, filtered rows only.
const small = t.clone(t.filter, ['smiles', 'activity']);
return DG.Viewer.grid(small).root;

col.clone(mask?) clones a single column, but the cloned column is not attached to any DataFrame. Call df.columns.add(clonedCol) to attach it.

Adding columns

You have...	Right call
Nothing — want empty typed column	t.columns.addNewFloat(name) (and addNewInt/String/Bool/...)
Generic typed empty column	t.columns.addNew(name, DG.COLUMN_TYPE.FLOAT)
An array of values	t.columns.add(DG.Column.fromList(DG.COLUMN_TYPE.FLOAT, name, values))
A function of row index	t.columns.addNewFloat(name).init(i => ...)
A formula using other columns	See datagrok-calc-column skill
Compute-on-demand (no storage)	t.columns.addNewVirtual(name, i => ..., DG.TYPE.STRING)
Want a collision-free name first	const n = t.columns.getUnusedName(name); then addNew*
Insert at a specific position	t.columns.insert(col, index)

DG.COLUMN_TYPE.FLOAT === 'double'. Never write the literal 'float' (no such type) or 'double' (works but brittle). Use the constant, or the typed shorthand addNewFloat.

// Typed empty column with metadata applied in the same block.
const col = t.columns.addNewFloat('Ki');
col.semType = DG.SEMTYPE.Ki;
col.meta.units = 'nM';
col.meta.format = '0.00';
return ui.divText(`Added ${col.name} (${col.type})`);

// Add a column from an array of values. fromList infers length from the array.
const values = [1.2, 3.4, null, 5.6];
const col = DG.Column.fromList(DG.COLUMN_TYPE.FLOAT, 'score', values);
t.columns.add(col);
return ui.divText(`Added ${col.name}`);

// Virtual column — recomputed on every access, not stored. Type argument is
// DG.TYPE.* (not DG.COLUMN_TYPE.*) — addNewVirtual takes the broader enum.
const labels = t.columns.addNewVirtual('Label',
  (i) => `${t.get('compound', i)}@${t.get('target', i)}`,
  DG.TYPE.STRING);
return ui.divText(`Added virtual column ${labels.name}`);

// Avoid name collisions explicitly.
const name = t.columns.getUnusedName('Ki');  // returns 'Ki' or 'Ki (1)' etc.
const col = t.columns.addNewFloat(name);
return ui.divText(`Final name: ${col.name}`);

Removing & renaming

// Remove one column. Accepts string, index, or Column instance.
t.columns.remove('temp');

// Remove several. Loop — no bulk remove call exists.
for (const name of ['_tmp1', '_tmp2', 'temp3'])
  if (t.columns.contains(name))
    t.columns.remove(name);

To target calculated columns, check col.meta.formula:

// Drop every calculated column.
const names = t.columns.toList().filter((c) => c.meta.formula != null).map((c) => c.name);
for (const n of names) t.columns.remove(n);
return ui.divText(`Removed: ${names.join(', ') || '(none)'}`);

Rename: assign col.name directly.

t.getCol('pIC50').name = 'pIC50_old';

Setting column metadata

Use the right accessor — never setTag for these.

Where metadata lives	Read	Write
Semantic type	col.semType	col.semType = DG.SEMTYPE.MOLECULE
Friendly name	col.meta.friendlyName	col.meta.friendlyName = 'Potency' (or null to clear)
Description	col.meta.description	col.meta.description = '...'
Format	col.meta.format	col.meta.format = '0.00'
Units	col.meta.units	col.meta.units = 'nM'
Choices (combo)	col.meta.choices	col.meta.choices = ['A', 'B']
Color coding	col.meta.colors.getType()	col.meta.colors.setLinear/setCategorical/setConditional/setDisabled
Arbitrary tags	col.getTag(k)	col.setTag(k, v), col.tags.delete(k)
Layout-stable id	col.layoutColumnId	col.layoutColumnId = '...'

Assign null to a col.meta.* accessor to clear the underlying tag.

Trap. The semType tag key is 'quality' in storage. Don't use col.setTag('semType', ...) or col.setTag('quality', ...). Use the col.semType accessor — it sets the right tag and fires the right event. Friendly name / format / units / description go through col.meta.*, not through setTag.

// Set several metadata fields at once.
const col = t.getCol('Ki');
col.semType = DG.SEMTYPE.Ki;
col.meta.units = 'nM';
col.meta.format = '0.00';
col.meta.description = null;  // null clears

// Clear all custom (user-facing) metadata. Do NOT call col.tags.clear() —
// that also nukes system tags like .color-coding-type.
const col = t.getCol('activity');
col.meta.friendlyName = null;
col.meta.units = null;
col.meta.format = null;
col.meta.description = null;

Color coding

When the user doesn't name specific colors or rules, call the setter with no arguments — the platform's default scheme applies. Never invent a palette the user didn't ask for.

Lives under col.meta.colors. Four shapes:

Linear (numeric only)

// Default: no palette — the platform's standard linear scheme is used.
t.getCol('weight').meta.colors.setLinear();

// User asked for red → yellow → green, with min/max pinned.
t.getCol('age').meta.colors.setLinear(
  ['#ff0000', '#ffff00', '#00ff00'],
  {min: 19, max: 70},
);

Hex strings and ARGB integers both work in range. Optional belowMinColor / aboveMaxColor paint out-of-range values.

Linear on a non-numeric column is a logic error — the setter itself does not throw; guard with col.isNumerical if uncertain.

Categorical

// Default: no map — every category gets a color from the platform palette.
t.getCol('race').meta.colors.setCategorical();

// User asked for specific colors per category.
t.getCol('race').meta.colors.setCategorical(
  {'Asian': '#0000FF', 'Black': '#FF0000'},
  {fallbackColor: '#CCCCCC'},
);

Conditional (range-based rules)

// Default: no rules — the platform bins the values into 4 ranges automatically.
t.getCol('height').meta.colors.setConditional();

// User specified the thresholds. '20-170' means "value in the range 20..170".
// '<100' / '>50' also work.
t.getCol('height').meta.colors.setConditional({
  '20-170': '#00FF00',
  '170-190': '#220505',
});

Off

t.getCol('activity').meta.colors.setDisabled();

Trap. Don't use col.setTag('.color-coding-type', ...) to enable / disable coloring. It bypasses rule validation and the companion tag writes that setLinear / setCategorical / setConditional perform, producing broken color coding.

Working with cell values

Bulk init: col.init(scalar | (i) => value) is the right shape for an already-allocated column.

const col = t.columns.addNewInt('rowSquared');
col.init((i) => i * i);

For hot paths, mutate values with notify=false and call col.fireValuesChanged() once at the end:

const src = t.getCol('input');
const dst = t.getCol('output');
for (let i = 0; i < t.rowCount; i++)
  dst.set(i, src.get(i) * 2, false);
dst.fireValuesChanged();

Constants — cite, never invent

Constant family	Members
DG.COLUMN_TYPE	STRING='string', INT='int', FLOAT='double' (!), BOOL='bool', BYTE_ARRAY='byte_array', DATE_TIME='datetime', BIG_INT='bigint', QNUM='qnum', DATA_FRAME='dataframe', OBJECT='object'
DG.SEMTYPE	MOLECULE, MACROMOLECULE, MOLECULE3D, IC50, EC50, Ki, CONCENTRATION, VOLUME, EMAIL, URL, LATITUDE, LONGITUDE, IMAGE, FILE, CHEMICAL_REACTION, ...
DG.UNITS.Molecule	SMILES, MOLBLOCK, V3K_MOLBLOCK, INCHI
DG.COLOR_CODING_TYPE	CATEGORICAL, CONDITIONAL, LINEAR, OFF

Anti-patterns — and the fix

1. Looping t.rows to read / write column data. Slow and notification-heavy. Use t.getCol(name) and an index loop with set(..., false) plus a final col.fireValuesChanged().
2. col.setTag('color-coding-type', ...) / col.setTag('.color-coding-type', ...). Use col.meta.colors.setLinear/setCategorical/setConditional/setDisabled.
3. col.setTag('semType', 'Molecule'). Use col.semType = DG.SEMTYPE.MOLECULE.
4. col.setTag('friendlyName', '...'), col.setTag('format', '...'). Go through col.meta.friendlyName / col.meta.format.
5. 'float' or 'double' string literals for column type. Always DG.COLUMN_TYPE.FLOAT (or just addNewFloat).
6. t.copy(). Doesn't exist. Use t.clone().
7. col.toList().length for row count. Use col.length, or col.stats.valueCount for non-null count.
8. new Set(col.toList()).size for unique count. Use col.stats.uniqueCount. For string cols, col.categories.length also works.
9. Mutating col.getRawData() and forgetting to fire. End with col.fireValuesChanged() — otherwise viewers / grid don't refresh.
10. Forgetting await on t.columns.addNewCalculated. It's the only addNew* that returns a Promise<Column>. See datagrok-calc-column.
11. t.columns.byName(name) + if (col == null) when you want an error. Use t.getCol(name) — it throws on missing.
12. col.tags.clear() to "reset metadata". Nukes system tags too (color coding, choices, etc). Clear specific col.meta.* fields with null.