adapting the reader to the dataset

DESI.py

@@ -1,7 +1,8 @@
 import os
 import numpy as np
 import datasets
-from datasets import Features, Value, Array2D
+import pyarrow as pa
+from datasets import Features, Value, Array2D, Sequence
 from datasets.utils.logging import get_logger
 
 # TODO: Add BibTeX citation
@@ -29,23 +30,47 @@ _LICENSE = ""
 
 # Download URLs for different variants of the dataset
 # TODO: these files should be versionned
-_DATA_PATH = "{data_dir}/{sample}/{healpixel}.parquet"
-
+_DATA_PATH = "{sample}/{healpixel}.parquet"
+_HEALPIXELS = ['9144', '9145', '9146', '9147', '9148', '9149', '9150', '9151', '9334', '9335', '9337', '9338', '9339', '9340', '9341', '9342', 
+               '9343', '9424', '9425', '9427', '9428', '9429', '9430', '9431', '9512', '9514', '9600', '9791', '9812', '9813', '9814', '9815', 
+               '9817', '9818', '9819', '9820', '9821', '9822', '9823', '9829', '9830', '9831', '9834', '9835', '9836', '9837', '9838', '9839', 
+               '9840', '9841', '9842', '9843', '9844', '9845', '9848', '9849', '9850', '9851', '9877', '9878', '9879', '9884', '9885', '9887', 
+               '9920', '9921', '9922', '9923', '9924', '9925', '9926', '9928', '9929', '9930', '9931', '9936', '9983', '9984', '9986', '9987', 
+               '9992', '9993', '9994', '10016', '10145', '10146', '10147', '10148', '10150', '10151', '10152', '10153', '10154', '10155', '10156', 
+               '10157', '10158', '10159', '10197', '10198', '10199', '10204', '10205', '10206', '10207', '10228', '10229', '10230', '10231', '10236',
+               '10237', '10238', '10239', '10368', '10370', '10376', '10377', '10378', '10379', '10380', '10382', '10383', '10400', '10401', '10402', 
+               '10403', '10404', '10405', '10406', '10408', '10409', '10512', '10513', '10514', '10515', '10516', '10517', '10518', '10519', '10521', 
+               '10524', '10719', '10741', '10743', '11215', '11224', '11225', '11226', '11227', '11228', '11229', '11230', '11231', '11237', '11239', 
+               '11248', '11249', '11250', '11251', '11252', '11253', '11254', '11349', '11400', '11401', '11402', '11403', '11404', '11405', '11406', 
+               '11407', '11424', '11425', '11426', '11427', '11428', '11429', '11430', '11520', '11521', '11522', '11523', '11524', '11526', '11604', 
+               '11605', '11606', '11607', '11612', '11613', '11912', '11914', '11936', '15338', '15339', '15340', '15341', '15342', '15343', '15352', 
+               '15353', '15354', '15355', '15356', '15357', '15358', '15359', '16040', '16041', '16042', '16043', '16046', '25595', '25596', '25597', 
+               '25598', '25599', '25909', '25911', '25913', '25914', '25915', '25916', '25917', '25918', '25919', '25926', '25927', '25929', '25930', 
+               '25931', '25932', '25933', '25934', '25935', '25938', '25944', '25945', '25946', '25947', '25952', '25953', '25954', '25955', '25956', 
+               '25957', '25958', '25959', '25960', '25961', '25962', '25963', '25964', '25965', '25966', '25967', '25968', '25969', '25970', '25971', 
+               '25974', '25976', '25977', '25978', '25979', '25980', '25981', '25982', '25983', '26000', '26001', '26003', '26004', '26005', '26006', 
+               '26007', '26012', '26013', '26039', '26045', '26047', '26048', '26050', '26053', '26064', '26065', '26068', '26081', '26082', '26083', 
+               '26086', '26087', '26088', '26089', '26090', '26091', '26092', '26093', '26094', '26095', '26251', '26254', '26272', '26273', '26274', 
+               '26275', '26276', '26277', '26278', '26279', '26280', '26281', '26282', '26283', '26284', '26285', '26286', '26287', '26290', '26296', 
+               '26432', '26433', '26436', '26437', '26961', '26964', '26965', '26966', '26967', '26973', '27238', '27239', '27244', '27245', '27246', 
+               '27247', '27250', '27251', '27256', '27257', '27258', '27259', '27260', '27262', '27333', '27344', '27345', '27346', '27347', '27348',
+               '27648', '27649', '27650', '27651', '27656', '28027', '28030', '28031', '28112', '28113', '28114', '28115', '28116', '28117', '28118', 
+               '28119', '28120', '28121', '28123', '28124', '28125', '28126', '28127', '28149', '28151']
 _VERSION = "0.0.1"
 
 logger = get_logger(__name__)
 
-class DESI(datasets.GeneratorBasedBuilder):
+class DESI(datasets.ArrowBasedBuilder):
     """TODO: Short description of my dataset."""
 
     VERSION = _VERSION
 
     BUILDER_CONFIGS = [
-        datasets.BuilderConfig(name="edr_sv3", version=VERSION, 
+        datasets.BuilderConfig(name="sv3", version=VERSION, 
                                description="One percent survey from the DESI Early Data Release."),
     ]
 
-    DEFAULT_CONFIG_NAME = "edr_sv3"
+    DEFAULT_CONFIG_NAME = "sv3"
 
     def _info(self):
         """ Defines the features available in this dataset.
@@ -54,14 +79,25 @@ class DESI(datasets.GeneratorBasedBuilder):
             # This is the description that will appear on the datasets page.
             description=_DESCRIPTION,
             # This defines the different columns of the dataset and their types
-            features=Features({
-                'spectrum': Array2D(shape=(None, 2), dtype='float32'), # Stores flux and ivar
-                'lambda_min': Value('float32'), # Min and max wavelength
-                'lambda_max': Value('float32'),
-                'resolution': Value('float32'), # Resolution of the spectrum
-                'z': Value('float32'),
-                'ebv': Value('float32'),
-            }),
+            features=Features(
+                {'TARGETID': Value(dtype='int64'),
+                'HEALPIX': Value(dtype='int32'),
+                'TARGET_RA': Value(dtype='float32'),
+                'TARGET_DEC': Value(dtype='float32'),
+                'RELEASE': Value(dtype='int16'),
+                'BRICKID': Value(dtype='int32'),
+                'BRICK_OBJID': Value(dtype='int32'),
+                'Z': Value(dtype='float32'),
+                'EBV': Value(dtype='float32'),
+                'FLUX_G': Value(dtype='float32'),
+                'FLUX_R': Value(dtype='float32'),
+                'FLUX_Z': Value(dtype='float32'),
+                'FLUX_IVAR_G': Value(dtype='float32'),
+                'FLUX_IVAR_R': Value(dtype='float32'),
+                'FLUX_IVAR_Z': Value(dtype='float32'),
+                'SPECTRUM': Sequence(feature=Value(dtype='float32'), length=7781),
+                'SPECTRUM_IVAR': Sequence(feature=Value(dtype='float32'), length=7781)
+                }),
             # Homepage of the dataset for documentation
             homepage=_HOMEPAGE,
             # License for the dataset if available
@@ -71,61 +107,16 @@ class DESI(datasets.GeneratorBasedBuilder):
         )
 
     def _split_generators(self, dl_manager):
-        # First, attempt to access the files locally, if unsuccessful, emit a warning and attempt to download them
-        if dl_manager.manual_dir is not None:
-            data_dir = dl_manager.manual_dir
-            data_dir = {k: os.path.join(data_dir, self.URLS[k].split('/')[-1]) 
-                        for k in self.URLS}
-        else:
-            logger.warning("We recommend downloading data manually through GLOBUS" 
-                           "and specifying the manual_dir argument to pass to the dataset builder."
-                           "Downloading data automatically through the dataset builder will proceed but is not recommended.")
-            data_dir = dl_manager.download_and_extract(self.URLS)
-
+        data_urls = [ _DATA_PATH.format(sample=self.config.name, healpixel=hp) for hp in _HEALPIXELS ]
+        downloaded_files = dl_manager.download_and_extract(data_urls)
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
-                gen_kwargs={**data_dir}
+                gen_kwargs={"filepaths": downloaded_files}
             )
         ]
-
-    def _generate_examples(self, catalog, data, keys = None):
-        """ Yields examples as (key, example) tuples.
-        """
-        import h5py
-        from astropy.table import Table
-
-        # Opening the catalog
-        catalog = Table.read(catalog)
-
-        # If no keys are provided, return all the examples
-        if keys is None:
-            keys = catalog['TARGETID']
-
-        # Preparing an index for fast searching through the catalog
-        sort_index = np.argsort(catalog['TARGETID'])
-        sorted_ids = catalog['TARGETID'][sort_index]
-
-        # Opening data file and iterating over the requested keys
-        with h5py.File(data, 'r') as data:
-            # Loop over the indices and yield the requested data
-            for i, id in enumerate(keys):
-                # Extract the indices of requested ids in the catalog 
-                idx = sort_index[np.searchsorted(sorted_ids, id)]
-                row = catalog[idx]
-                key = row['TARGETID']
-
-                example = {
-                    'spectrum':  np.stack([data['flux'][idx],
-                                                    data['ivar'][idx]], axis=1).astype('float32'),# TODO: add correct values
-                    'lambda_min':  0.,  # TODO: add correct values
-                    'lambda_max':  1.,  # TODO: add correct values
-                    'resolution':  0.1,
-                    'z':  row['Z'],
-                    'ebv':  row['EBV'],
-                }
-
-                # Checking that we are retriving the correct data
-                assert (key == keys[i]) & (data['target_ids'][idx] == keys[i]) , ("There was an indexing error when reading desi spectra", (key, keys[i]))
-
-                yield str(key), example
+    
+    def _generate_tables(self, filepaths):
+        for hp, filepath in zip(_HEALPIXELS, filepaths):
+            table = pa.parquet.read_table(filepath)
+            yield hp, table