Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Create SplitMatrix from polars data frame #329

Closed
jtilly opened this issue Nov 6, 2023 · 0 comments · Fixed by #370
Closed

Create SplitMatrix from polars data frame #329

jtilly opened this issue Nov 6, 2023 · 0 comments · Fixed by #370

Comments

@jtilly
Copy link
Member

jtilly commented Nov 6, 2023

Currently, we have a handy from_pandas constructor

def from_pandas(
df: pd.DataFrame,
dtype: np.dtype = np.float64,
sparse_threshold: float = 0.1,
cat_threshold: int = 4,
object_as_cat: bool = False,
cat_position: str = "expand",
drop_first: bool = False,
) -> MatrixBase:
"""
Transform a pandas.DataFrame into an efficient SplitMatrix. For most users, this
will be the primary way to construct tabmat objects from their data.
Parameters
----------
df : pd.DataFrame
pandas DataFrame to be converted.
dtype : np.dtype, default np.float64
dtype of all sub-matrices of the resulting SplitMatrix.
sparse_threshold : float, default 0.1
Density threshold below which numerical columns will be stored in a sparse
format.
cat_threshold : int, default 4
Number of levels of a categorical column under which the column will be stored
as sparse one-hot-encoded columns instead of CategoricalMatrix
object_as_cat : bool, default False
If True, DataFrame columns stored as python objects will be treated as
categorical columns.
cat_position : str {'end'|'expand'}, default 'expand'
Position of the categorical variable in the index. If "last", all the
categoricals (including the ones that did not satisfy cat_threshold)
will be placed at the end of the index list. If "expand", all the variables
will remain in the same order.
drop_first : bool, default False
If true, categoricals variables will have their first category dropped.
This allows multiple categorical variables to be included in an
unregularized model. If False, all categories are included.
Returns
-------
SplitMatrix
"""
matrices: list[Union[DenseMatrix, SparseMatrix, CategoricalMatrix]] = []
indices: list[list[int]] = []
is_cat: list[bool] = []
dense_dfidx = [] # column index in original DataFrame
dense_mxidx = [] # index in the new SplitMatrix
sparse_dfcols = [] # sparse columns to join together
sparse_mxidx = [] # index in the new SplitMatrix
ignored_cols = []
mxcolidx = 0
for dfcolidx, (colname, coldata) in enumerate(df.items()):
# categorical
if object_as_cat and coldata.dtype == object:
coldata = coldata.astype("category")
if isinstance(coldata.dtype, pd.CategoricalDtype):
cat = CategoricalMatrix(coldata, drop_first=drop_first, dtype=dtype)
if len(coldata.cat.categories) < cat_threshold:
(
X_dense_F,
X_sparse,
dense_indices,
sparse_indices,
) = _split_sparse_and_dense_parts(
sps.csc_matrix(cat.tocsr(), dtype=dtype),
threshold=sparse_threshold,
)
matrices.append(X_dense_F)
is_cat.append(True)
matrices.append(X_sparse)
is_cat.append(True)
if cat_position == "expand":
indices.append(mxcolidx dense_indices)
indices.append(mxcolidx sparse_indices)
mxcolidx = len(dense_indices) len(sparse_indices)
elif cat_position == "end":
indices.append(dense_indices)
indices.append(sparse_indices)
else:
matrices.append(cat)
is_cat.append(True)
if cat_position == "expand":
indices.append(mxcolidx np.arange(cat.shape[1]))
mxcolidx = cat.shape[1]
elif cat_position == "end":
indices.append(np.arange(cat.shape[1]))
# All other numerical dtypes (needs to be after pd.SparseDtype)
elif is_numeric_dtype(coldata):
# check if we want to store as sparse
if (coldata != 0).mean() <= sparse_threshold:
if not isinstance(coldata.dtype, pd.SparseDtype):
sparse_dtype = pd.SparseDtype(coldata.dtype, fill_value=0)
sparse_dfcols.append(coldata.astype(sparse_dtype))
else:
sparse_dfcols.append(coldata)
sparse_mxidx.append(mxcolidx)
mxcolidx = 1
else:
dense_dfidx.append(dfcolidx)
dense_mxidx.append(mxcolidx)
mxcolidx = 1
# dtype not handled yet
else:
ignored_cols.append((dfcolidx, colname))
if len(ignored_cols) > 0:
warnings.warn(
f"Columns {ignored_cols} were ignored. Make sure they have a valid dtype."
)
if len(dense_dfidx) > 0:
matrices.append(DenseMatrix(df.iloc[:, dense_dfidx].astype(dtype)))
indices.append(dense_mxidx)
is_cat.append(False)
if len(sparse_dfcols) > 0:
sparse_dict = {i: v for i, v in enumerate(sparse_dfcols)}
full_sparse = pd.DataFrame(sparse_dict).sparse.to_coo()
matrices.append(SparseMatrix(full_sparse, dtype=dtype))
indices.append(sparse_mxidx)
is_cat.append(False)
if cat_position == "end":
new_indices = []
for mat_indices, is_cat_ in zip(indices, is_cat):
if is_cat_:
new_indices.append(np.asarray(mat_indices) mxcolidx)
mxcolidx = len(mat_indices)
else:
new_indices.append(mat_indices)
indices = new_indices
if len(matrices) > 1:
return SplitMatrix(matrices, indices)
elif len(matrices) == 0:
raise ValueError("DataFrame contained no valid column")
else:
return matrices[0]

that constructs a SplitMatrix from a pandas data frame.

It would great to also support polars (or arrow tables more generally given that it's easy to move from polars to arrow). This would open the door to having glum support polars data frames as input.

As a reference, @borchero just added arrow support to LightGBM: microsoft/LightGBM#6034

@MarcAntoineSchmidtQC MarcAntoineSchmidtQC linked a pull request Jun 18, 2024 that will close this issue
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

1 participant