pandas/chained-assignment
| Rule | chained-assignment |
|---|---|
| Pack | pandas |
| Severity | error |
| Fix | unsafe, with xpd check --fix --unsafe-fixes |
| Library versions | every version |
Under copy-on-write, assigning through a chain such as df["a"][i] = v writes to a copy, so the frame keeps its old values. Assign to the frame in one step with .loc.
df["load"][row] = valuedf.loc[row, "load"] = valueWhat the rule reports
Section titled “What the rule reports”The rule reports df["a"][i] = v, df[mask]["a"] = v, df.loc[mask]["a"] = v, df.a[mask] = v, df["a"].iloc[0] = v and df["a"].update(other), which write to a copy, so the frame keeps its old values. It also reports df[col][mask] = v, where a variable names the column.
The rule leaves alone code that more often fills a list or a dict, as the table shows.
| Left alone | Reason |
|---|---|
a dict, list or set assigned to x["a"][key] |
the code fills a nested dict |
two plain keys, as in rows[0]["name"] = v |
the value is more often a list or a dict |
a name that the file shows holds a dict, as in tables["direction"].loc[row] = v after tables = {...} |
the first key takes a frame or an array out of the dict, and a dict has no .loc |
How to fix it
Section titled “How to fix it”xpd check --fix --unsafe-fixes applies the fixes of this rule. Check the result, because an unsafe fix can change what the code does.
The unsafe fix writes df.loc[i, "a"] = v, and df.loc[df.index[0], "a"] = v for a write through .iloc or .iat on a column, which writes every row when the index holds the label twice. Other chains and update on a column have no fix.
The fix is unsafe, because xpd cannot always see whether the value on the left is a frame. A name
counts as a dict when the function or the module assigns it a dict, a dict comprehension or a call
of dict, or annotates it as dict or Mapping. A dict that a parameter without an annotation or
another function's result holds is still reported. The table shows which forms get a fix.
| Reported form | Fix |
|---|---|
df["a"][i] = v and df[col][mask] = v |
df.loc[i, "a"] = v and df.loc[mask, col] = v |
df[mask]["a"] = v |
df.loc[mask, "a"] = v |
df.loc[mask]["a"] = v |
df.loc[mask, "a"] = v |
any other chain with a text label, such as df.iloc[0]["a"] = v |
none |
df["a"].iloc[0] = v and df["a"].iat[0] = v |
df.loc[df.index[0], "a"] = v |
df.a[mask] = v and df["a"].update(other) |
none |
A mask held in a name gets the fix when the enclosing function assigns only masks to the name, such
as a comparison or a call to isin. The rule reports df.a[mask] = v only when the mask reads the
same frame, so self.items[i > 3] = v is left alone. It reports update only on a name such as
df or frame, because state["seen"].update(extra) is how code updates a dict of dicts.
Message
Section titled “Message”xpd prints this message with the issue. A name that starts with $ stands for code from your file.
Chained assignment writes to a copy under copy-on-write, so the frame keeps its old values. Assign to $DF.loc[$ROW, $COL] instead.Browse the other rules of the pandas pack.