Always treat price data as pd.DataFrame, never as pd.Series

[andreas-vester]

I realized that vectorbt is treating price data differently depending on the number of symbols provided.

When providing a single symbol, vectorbt is returning a pd.Series, while it returns a pd.DataFrame when providign multiple symbols.

The following get function is responsible for this distinction.

vectorbt/vectorbt/data/base.py

Lines 699 to 718 in f78d1fd

	def get(self, column: tp.Optional[tp.Label] = None, **kwargs) -> tp.MaybeTuple[tp.SeriesFrame]:
	"""Get column data.
	If one symbol, returns data for that symbol.
	If multiple symbols, performs concatenation first and returns a DataFrame if one column
	and a tuple of DataFrames if a list of columns passed."""
	if len(self.symbols) == 1:
	if column is None:
	return self.data[self.symbols[0]]
	return self.data[self.symbols[0]][column]
	concat_data = self.concat(**kwargs)
	if len(concat_data) == 1:
	return tuple(concat_data.values())[0]
	if column is not None:
	if isinstance(column, list):
	return tuple([concat_data[c] for c in column])
	return concat_data[column]
	return tuple(concat_data.values())

It might look reasonable to treat a single column as a pd.Series, however, this will possibly create problems/issues in downstream code.

For example, I m further processing the data that I retrive when applying the get method in my code. I need to ALWAYS get back a pd.DataFrame with the same column level names no matter how many symbols I provide.

There MIGHT be a quick solution for this issue. I changed part of the function in the following way:

def get(self, column: tp.Optional[tp.Label] = None, **kwargs) -> tp.MaybeTuple[tp.SeriesFrame]:
    """Get column data.

    If one symbol, returns data for that symbol.
    If multiple symbols, performs concatenation first and returns a DataFrame if one column
    and a tuple of DataFrames if a list of columns passed."""
    if len(self.symbols) == 1:
        if column is None:
            return self.data[self.symbols[0]]  # maybe this part needs to be adjusted, too?

        # NEW CODE
        df = pd.DataFrame(self.data[self.symbols[0]][column])
        df = df.rename({column: self.symbols[0]}, axis=1)
        df.columns = df.columns.set_names("symbol")

        return df

With this small change my downstream code seems to work properly, no matter if I provide a single or multiple symbols (fingers crossed).

Having said that, I would be more confident, if you were to change and test the library code.

In my experience, mixing pd.Series and pd.DataFrame causes trouble more often than not. I always go with pd.DataFrame no matter how many columns.

[polakowo]

I agree fully, there's an entire argument in the pro version called single_symbol that switches this behavior but there's no such argument in the open-source version. Maybe here we could add an argument return_frame in Data.get, I'll add it to feature requests.

[github-actions]

This issue is stale because it has been open for 90 days with no activity.

[andreas-vester]

@polakowo Any chance that issue will be solved in the near future/at all?