pathwaycom/pathway · error · ValueError

statistical.diff(): Invalid column reference for the paramet

Error message

statistical.diff(): Invalid column reference for the parameter timestamp, found a string. Did you mean this.{timestamp} instead of {repr(timestamp)}?

What it means

statistical.diff() computes the difference of each value column with respect to the previous row, ordered by a timestamp column. The timestamp parameter must be a pw.ColumnReference (or something resolvable through self[...]); passing a bare Python string is almost always the mistake of giving a column name instead of the column itself, so the method raises this ValueError and the message explicitly suggests the this.<name> reference form.

Source

Thrown at python/pathway/stdlib/ordered/diff.py:89

    ... 6         | 0        | 16
    ... '''
    ... )
    >>> table += table.diff(pw.this.timestamp, pw.this.values, instance=pw.this.instance)
    >>> pw.debug.compute_and_print(table, include_id=False)
    timestamp | instance | values | diff_values
    1         | 0        | 1      |
    2         | 1        | 2      |
    3         | 0        | 7      | 6
    3         | 1        | 4      | 2
    6         | 0        | 16     | 9
    6         | 1        | 11     | 7
    """

    if isinstance(timestamp, pw.ColumnReference):
        timestamp = self[timestamp]
    else:
        if isinstance(timestamp, str):
            raise ValueError(
                "statistical.diff(): Invalid column reference for the parameter timestamp,"
                + f" found a string. Did you mean this.{timestamp} instead of {repr(timestamp)}?"
            )
        raise ValueError(
            "statistical.diff(): Invalid column reference for the parameter timestamp."
        )

    ordered_table = self.sort(key=timestamp, instance=instance)

    for value in values:
        if isinstance(value, pw.ColumnReference):
            value = self[value]
        else:
            if isinstance(value, str):
                raise ValueError(
                    "statistical.diff(): Invalid column reference for the parameter value,"
                    + f" found a string. Did you mean this.{value} instead of {repr(value)}?"
                )

View on GitHub (pinned to fa2f74a464)

Solutions

Pass a column reference: timestamp=pw.this.ts (or table.ts).
If building calls dynamically from names, resolve the reference first: timestamp=getattr(pw.this, name).

Example fix

# before
t.statistical.diff(timestamp="ts", values=[pw.this.val])

# after
t.statistical.diff(timestamp=pw.this.ts, values=[pw.this.val])

Defensive patterns

Strategy: type-guard

Type guard

import pathway as pw

def as_column_reference(table, timestamp):
    if isinstance(timestamp, str):
        return table[timestamp]  # or getattr(pw.this, timestamp)
    if isinstance(timestamp, pw.ColumnReference):
        return timestamp
    raise TypeError("timestamp must be a pw.ColumnReference")

Try / catch

try:
    t.statistical.diff(timestamp=ts, values=[pw.this.v])
except ValueError as e:
    if "found a string" in str(e):
        t.statistical.diff(timestamp=pw.this[ts], values=[pw.this.v])
    else:
        raise

Prevention

Always pass pw.this.col references, not names, to Pathway APIs that take ColumnReference parameters.
When generating calls from schema metadata, resolve names to references with getattr(pw.this, name) at the call boundary.

When it happens

Trigger: Calling table.statistical.diff(timestamp="ts", values=[...]) — i.e. a string column name — instead of timestamp=pw.this.ts. Any str value hits the string-specific branch with the did-you-mean hint; other invalid types raise the generic variant.

Common situations: Habit from pandas/Spark APIs where column names are passed as strings; dynamic code building kwargs from schema column-name lists; refactoring from methods that accept both names and references.

Related errors

AI-assisted analysis of pathwaycom/pathway@fa2f74a464 (2026-08-15). Data as JSON: /api/errors/02148ce77be20bcc. Report an issue: GitHub.