Wei Ji
commited on
Initial script to load embeddings from geoparquet file
Browse filesExperimenting with ArrowBasedBuilder, mostly with help from https://discuss.huggingface.co/t/how-to-tweak-a-dataset-without-a-loading-script/43533/5
- clay_vector_embeddings.py +70 -0
clay_vector_embeddings.py
ADDED
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Custom HuggingFace dataset loading script for GeoParquet files.
|
3 |
+
|
4 |
+
References:
|
5 |
+
- https://huggingface.co/docs/datasets/v2.15.0/en/dataset_script
|
6 |
+
- https://github.com/huggingface/datasets/blob/2.15.0/templates/new_dataset_script.py
|
7 |
+
- https://huggingface.co/docs/datasets/v2.15.0/en/package_reference/builder_classes
|
8 |
+
- https://huggingface.co/docs/datasets/v2.15.0/en/about_dataset_load
|
9 |
+
- https://discuss.huggingface.co/t/how-to-tweak-a-dataset-without-a-loading-script/43533/5
|
10 |
+
"""
|
11 |
+
import datasets
|
12 |
+
import pyarrow as pa
|
13 |
+
import pyarrow.parquet as pq
|
14 |
+
|
15 |
+
|
16 |
+
_URLS = {"32VLM": " 32VLM_v01.gpq"}
|
17 |
+
_MGRS_TILES = ["32VLM"]
|
18 |
+
|
19 |
+
|
20 |
+
class ClayVectorEmbeddings(datasets.ArrowBasedBuilder):
|
21 |
+
"""Clay Vector Embeddings in GeoParquet format."""
|
22 |
+
|
23 |
+
# You will be able to load one or the other configurations in the following list with
|
24 |
+
# data = datasets.load_dataset('my_dataset', 'MGRS_TILE')
|
25 |
+
BUILDER_CONFIGS = [
|
26 |
+
datasets.BuilderConfig(
|
27 |
+
name=name,
|
28 |
+
version=datasets.Version(version="0.0.1"),
|
29 |
+
description=f"Clay vector embeddings for MGRS tile {name}",
|
30 |
+
)
|
31 |
+
for name in _MGRS_TILES
|
32 |
+
]
|
33 |
+
|
34 |
+
# DEFAULT_CONFIG_NAME = "32VLM"
|
35 |
+
|
36 |
+
def _info(self):
|
37 |
+
return datasets.DatasetInfo(
|
38 |
+
# This is the description that will appear on the datasets page.
|
39 |
+
description="Clay Vector Embeddings in GeoParquet format.",
|
40 |
+
# This defines the different columns of the dataset and their types
|
41 |
+
features=datasets.Features(
|
42 |
+
{
|
43 |
+
"source_url": datasets.Value(dtype="string"),
|
44 |
+
"date": datasets.Value(dtype="date32"),
|
45 |
+
"embeddings": datasets.Value("string"),
|
46 |
+
"geometry": datasets.Value("binary"),
|
47 |
+
# These are the features of your dataset like images, labels ...
|
48 |
+
}
|
49 |
+
),
|
50 |
+
)
|
51 |
+
|
52 |
+
def _split_generators(self, dl_manager: datasets.download.DownloadManager):
|
53 |
+
files = _URLS[self.config.name]
|
54 |
+
downloaded_files = dl_manager.download(files)
|
55 |
+
return [
|
56 |
+
datasets.SplitGenerator(
|
57 |
+
name=datasets.Split.ALL,
|
58 |
+
# These kwargs will be passed to _generate_tables
|
59 |
+
gen_kwargs={"filepaths": downloaded_files},
|
60 |
+
)
|
61 |
+
]
|
62 |
+
|
63 |
+
# method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
|
64 |
+
def _generate_tables(self, filepaths: list[str] = ["32VLM_v01.gpq"]):
|
65 |
+
for file_idx, filepath in enumerate(filepaths):
|
66 |
+
with open(filepath, mode="rb") as f:
|
67 |
+
parquet_file = pq.ParquetFile(source=filepath)
|
68 |
+
for batch_idx, record_batch in enumerate(parquet_file.iter_batches()):
|
69 |
+
pa_table = pa.Table.from_batches([record_batch])
|
70 |
+
yield f"{file_idx_batch_idx}", pa_table
|