datatable uses a unified DT[i, j] indexing syntax. i selects rows and j selects columns. Either can be omitted or set to : / ... to mean “all”.
from datatable import dt, f
from datetime import date
source = {
"dates": [date(2000, 1, 5), date(2010, 11, 23), date(2020, 2, 29), None],
"integers": range(1, 5),
"floats": [10.0, 11.5, 12.3, -13],
"strings": ['A', 'B', None, 'D'],
}
DT = dt.Frame(source)
Selecting Columns
The j position controls which columns are returned.
By name
DT[:, 'dates']
# or shorthand (all rows implied):
DT['dates']
By position
DT[..., 2] # 3rd column
DT[:, -2] # 2nd column from the end
DT[0] # shorthand — first column, all rows
By list of names or positions
DT[:, ['integers', 'strings']]
DT[:, (-3, 2, 3)] # tuple of positions
TypeError.
By slice
String slices are inclusive on both ends:
DT[:, 'dates':'strings'] # all four columns
DT[:, slice('integers', 'strings')]
DT[:, 1:3] # columns at index 1 and 2
DT[:, ::-1] # all columns reversed
DT[:, [slice("dates", "integers"), slice("floats", "strings")]]
By data type
DT[:, int] # all int columns
DT[:, dt.Type.float64] # exact type
DT[:, [date, str]] # list of types
By boolean mask
The mask length must equal the number of columns:
DT[:, [True, True, False, False]]
# List comprehension — select columns whose name contains "i"
DT[:, ["i" in name for name in DT.names]]
# Select numeric columns with mean > 3
DT[:, [col.type.is_numeric and col.mean1() > 3 for col in DT]]
Via f-expressions
The f symbol mirrors all column-selector options:
DT[:, f.dates]
DT[:, f[-1]] # last column
DT[:, f['integers':'strings']] # named slice
DT[:, f[date, int, float]] # multiple types
DT[:, f["dates":"integers", "floats":"strings"]] # multi-slice
If a column name is a Python keyword (def, del, …) use bracket notation f['del'] — dot access is not available.
Removing columns with .remove()
DT[:, f[:].remove(f.dates)] # remove by name
DT[:, f[:].remove(f[0])] # remove by position
DT[:, f[:].remove(f[1:3])] # remove by slice
DT[:, f[:].remove(f[int, float])] # remove by type
Filtering Rows
The i position controls which rows are returned.
By single position
DT[0, :] # first row
DT[-1, :] # last row
By sequence of positions
DT[[1, 2, 3], :] # list
DT[range(1, 3), :] # range
DT[dt.Frame([1, 2, 3]), :] # one-column integer Frame
By slice
DT[1:3, :] # rows 1 and 2
DT[::-1, :] # all rows reversed
DT[-1:-3:-1, :] # last two rows, reversed
# Multiple slices
DT[[slice(1, 3), slice(5, 8)], :]
By boolean sequence
DT[[True, True, False, False], :]
# Generator expression
DT[(n % 2 == 0 for n in range(DT.nrows)), :]
By f-expressions (boolean filter)
# Date comparison
DT[f.dates < dt.Frame([date(2020, 1, 1)]), :]
# Arithmetic condition
DT[f.integers % 2 != 0, :]
# Compound condition
DT[(f.integers == 3) & (f.strings == None), ...]
# Filter by type group
DT[f[float] < 1, :]
# Row-wise aggregation filter
DT[dt.rowsum(f[int, float]) > 12, :]
Combined Row and Column Selection
Both i and j can be specified together:
DT[0, slice(1, 3)] # first row, columns 1–2
DT[2:6, ["i" in name for name in DT.names]]
DT[f.integers > dt.mean(f.floats) - 3, f['strings':'integers']]
Single value access
Passing integer scalars to both i and j returns a Python scalar:
DT[0, 0] # datetime.date(2000, 1, 5)
DT[0, 2] # 10.0
DT[-3, 'strings'] # 'B'
Excluding (Deselecting) Rows and Columns
Use list comprehensions to exclude specific columns or rows without deleting them:
# Exclude one column
DT[:, [name for name in DT.names if name != "integers"]]
# Exclude multiple columns
DT[:, [name not in ("integers", "dates") for name in DT.names]]
# Exclude non-numeric columns
DT[2:7, [not coltype.is_numeric for coltype in DT.types]]
# Exclude a range of rows via complementary slices
DT[[slice(None, 3), slice(7, None)], :]
Deleting Rows and Columns
del performs in-place deletion — no reassignment needed.
# Delete multiple rows
del DT[3:7, :]
# Delete a single row
del DT[3, :]
# Delete a column by name
del DT['strings']
# Delete multiple columns
del DT[:, ['dates', 'floats']]
Deletion with del is destructive and in-place. There is no undo — copy the frame first if you need to preserve the original.