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.

Sign up for GitHub

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

Support partitioning and group size control in coco dataset generation. #175

Merged
merged 2 commits into from
Sep 19, 2022
Merged

Support partitioning and group size control in coco dataset generation. #175

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5248e6e
pass partition and group size control to coco dataset
eddyxu Sep 19, 2022
1ee62aa
be able to support home path
eddyxu Sep 19, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 4 additions & 3 deletions python/benchmarks/bench_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@

def read_file(uri) -> bytes:
if not urlparse(uri).scheme:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

are you allowed to use ~ substitution in file:// uri's?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yea, this way we can use "train.py ~/dataset".

uri = pathlib.Path(uri)
uri = pathlib.Path(uri).expanduser().absolute()
fs, key = pyarrow.fs.FileSystem.from_uri(uri)
return fs.open_input_file(key).read()

Expand Down Expand Up @@ -221,16 +221,17 @@ def default_dataset_path(self, fmt, flavor=None):
suffix = f"_{flavor}" if flavor else ""
return os.path.join(self.uri_root, f"{self.name}{suffix}.{fmt}")

def save_df(self, df, fmt="lance", output_path=None):
def save_df(self, df, fmt="lance", output_path=None, **kwargs):
output_path = output_path or self.default_dataset_path(fmt, "links")
table = self._convert_metadata_df(df)
if fmt == "parquet":
pq.write_table(table, output_path)
pq.write_table(table, output_path, **kwargs)
elif fmt == "lance":
pa.dataset.write_dataset(
table,
output_path,
format=lance.LanceFileFormat(),
**kwargs,
)
return table

Expand Down
58 changes: 42 additions & 16 deletions python/benchmarks/parse_coco.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,7 +65,7 @@ def read_split(split):
return self._instances_to_df(split, json_data)

df = pd.concat([read_split(split) for split in ["train", "val"]])
df['date_captured'] = pd.to_datetime(df.date_captured) # lance GH#98
df["date_captured"] = pd.to_datetime(df.date_captured) # lance GH#98
return df

def _convert_metadata_df(self, df: pd.DataFrame) -> pa.Table:
Expand All @@ -88,23 +88,22 @@ def _convert_field(self, name, typ, col):
offsets = native_arr.offsets
values = native_arr.values.to_numpy(zero_copy_only=False)
return pa.ListArray.from_arrays(
offsets, self._convert_field(
f'{name}.elements', typ.value_type, values)
offsets, self._convert_field(f"{name}.elements", typ.value_type, values)
)
elif pa.types.is_struct(typ):
native_arr = pa.array(col)
arrays = []
for subfield in typ:
sub_arr = native_arr.field(subfield.name)
if name == 'annotations' and subfield.name == 'name':
if name == "annotations" and subfield.name == "name":
converted = self._convert_name_column(
sub_arr, native_arr.field('category_id')
sub_arr, native_arr.field("category_id")
)
else:
converted = self._convert_field(
f"{name}.{subfield.name}",
subfield.type,
sub_arr.to_numpy(zero_copy_only=False)
sub_arr.to_numpy(zero_copy_only=False),
)
arrays.append(converted)
return pa.StructArray.from_arrays(arrays, fields=typ)
Expand All @@ -115,17 +114,21 @@ def _convert_field(self, name, typ, col):
def _convert_name_column(self, name_arr, category_id_arr):
coco_classes = pd.read_csv("coco_classes.csv", header=0, index_col=None)
# let's make sure the actual data matches
check = pd.Series(dict(zip(name_arr.values.to_numpy(False),
category_id_arr.values.to_numpy(False)))
).to_frame(name='check_id')
joined = coco_classes.set_index('name').join(check, how='right')
check = pd.Series(
dict(
zip(
name_arr.values.to_numpy(False),
category_id_arr.values.to_numpy(False),
)
)
).to_frame(name="check_id")
joined = coco_classes.set_index("name").join(check, how="right")
mask = pd.notnull(joined.check_id)
filtered = joined[mask]
if not (filtered.check_id == filtered.category_id).all():
raise ValueError(f"Category id check failed")
dict_arr = pa.DictionaryArray.from_pandas(
pd.Categorical(name_arr.values.to_numpy(False),
coco_classes.name.values)
pd.Categorical(name_arr.values.to_numpy(False), coco_classes.name.values)
)
assert not pd.isna(dict_arr.indices.to_numpy()).all()
return pa.ListArray.from_arrays(name_arr.offsets, dict_arr)
Expand Down Expand Up @@ -153,7 +156,7 @@ def get_schema(self):
ImageType.from_storage(pa.utf8()),
pa.int64(),
pa.int64(),
pa.timestamp('ns'),
pa.timestamp("ns"),
ImageType.from_storage(pa.utf8()),
pa.int64(),
pa.dictionary(pa.int8(), pa.utf8()),
Expand Down Expand Up @@ -218,13 +221,28 @@ def _aggregate_annotations(annotations):
)
@click.option("-f", "--fmt", type=str, help="Output format (parquet or lance)")
@click.option("-e", "--embedded", type=bool, default=True, help="Embed images")
@click.option(
"-g",
"--group-size",
type=int,
default=1024,
help="set the group size",
show_default=True,
)
@click.option(
"--max-rows-per-file",
type=int,
default=0,
help="set the max rows per file",
show_default=True,
)
@click.option(
"-o",
"--output-path",
type=str,
help="Output path. Default is {base_uri}/coco_links.{fmt}",
)
def main(base_uri, version, fmt, embedded, output_path):
def main(base_uri, version, fmt, embedded, output_path, group_size: int, max_rows_per_file: int):
converter = CocoConverter(base_uri, version=version)
df = converter.read_metadata()
known_formats = ["lance", "parquet"]
Expand All @@ -233,11 +251,19 @@ def main(base_uri, version, fmt, embedded, output_path):
fmt = [fmt]
else:
fmt = known_formats

kwargs = {
"existing_data_behavior": "overwrite_or_ignore",
"partitioning": ["split"],
"partitioning_flavor": "hive",
"max_rows_per_group": group_size,
"max_rows_per_file": max_rows_per_file,
}
for f in fmt:
if embedded:
converter.make_embedded_dataset(df, f, output_path)
converter.make_embedded_dataset(df, f, output_path, **kwargs)
else:
return converter.save_df(df, f, output_path)
return converter.save_df(df, f, output_path, **kwargs)


if __name__ == "__main__":
Expand Down