File size: 24,641 Bytes
89d437f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
"""
Amazing Crop Yield Dataset (ACYD) - Hugging Face Dataset Implementation

A comprehensive multi-country crop yield prediction dataset with weather, land surface,
and yield data for machine learning applications.

Usage:
    load_dataset(
        "notadib/ACYD",
        country="argentina",
        crop_type="corn",
        standardize=True,
        test_year=2020,
        n_train_years=10,
        n_past_years=5,
        trust_remote_code=True
    )

Output format:
    - weather: (seq_len, n_weather_vars)
    - land_surface: (seq_len, n_land_surface_vars)
    - land_surface_mask: (seq_len, n_land_surface_vars) [used cause ndvi starts from 1982 but rest from 1979]
    - soil: (soil_depths, n_soil_vars)
    - coords: (1,2)
    - years: (n_past_years + 1, 1)
    - y_past: (n_past_years, 1)
    - y: (1,)
"""

import datasets
import logging
import pandas as pd
import numpy as np
from tqdm import tqdm
import os

# Default values if constants module is not available
MAX_CONTEXT_LENGTH = 500
N_WEATHER_VARS = 5  # precipitation, reference_et, snow_lwe, solar_radiation, t2m_max
N_LAND_SURFACE_VARS = 3  # lai_high, lai_low, ndvi
N_SOIL_VARS = 8  # Different soil properties
SOIL_DEPTHS = 6  # Different depth layers
CROP_YIELD_STATS = {
    "soybean": {"mean": [], "std": []},
    "corn": {"mean": [], "std": []},
    "wheat": {"mean": [], "std": []},
    "sunflower": {"mean": [], "std": []},
}

# Valid parameter ranges (updated for Argentina data)
_CROP_TYPES = ["soybean", "corn", "wheat", "sunflower"]
_TEST_YEARS = list(range(1982, 2025))  # 1982-2024 based on available data
_N_TRAIN_YEARS = list(range(1, 31))  # 1-30 years
_N_PAST_YEARS = list(range(1, 11))  # 1-10 years


class CropYieldConfig(datasets.BuilderConfig):
    """Custom configuration for Crop Yield Dataset."""

    def __init__(
        self,
        crop_type: str = "soybean",
        test_year: int = 2018,
        n_train_years: int = 10,
        n_past_years: int = 5,
        data_dir: str = "./",
        standardize: bool = True,
        **kwargs,
    ):
        """
        Args:
            crop_type: Type of crop (soybean, corn, wheat, sunflower)
            test_year: Year to use for testing (1982-2024)
            n_train_years: Number of years to use for training (1-30)
            n_past_years: Number of past years to include in features (1-10)
            data_dir: Directory containing the data files
            standardize: Whether to standardize the data
            **kwargs: Additional arguments for BuilderConfig
        """
        # Validate parameters
        assert crop_type in _CROP_TYPES, f"Crop type must be one of {_CROP_TYPES}"
        assert test_year in _TEST_YEARS, f"Test year must be between 1982 and 2024"
        assert (
            n_train_years in _N_TRAIN_YEARS
        ), f"Training years must be between 1 and 30"
        assert n_past_years in _N_PAST_YEARS, f"Past years must be between 1 and 10"

        # Create descriptive config name following GitHub Code style
        std_str = "std" if standardize else "nostd"
        config_name = (
            f"{crop_type}-{test_year}-{n_train_years}-{n_past_years}-{std_str}"
        )

        super().__init__(
            name=config_name,
            **kwargs,
        )

        self.crop_type = crop_type
        self.test_year = test_year
        self.n_train_years = n_train_years
        self.n_past_years = n_past_years
        self.data_dir = data_dir
        self.standardize = standardize


class CropYieldDataset(datasets.GeneratorBasedBuilder):
    """Crop Yield Dataset with weather and land surface data."""

    VERSION = datasets.Version("1.0.0")
    BUILDER_CONFIG_CLASS = CropYieldConfig
    config: CropYieldConfig  # Type annotation for config

    BUILDER_CONFIGS = [
        CropYieldConfig(
            crop_type="soybean",
            description="Soybean yield prediction dataset with default parameters",
        ),
        CropYieldConfig(
            crop_type="corn",
            description="Corn yield prediction dataset with default parameters",
        ),
        CropYieldConfig(
            crop_type="wheat",
            description="Wheat yield prediction dataset with default parameters",
        ),
        CropYieldConfig(
            crop_type="sunflower",
            description="Sunflower yield prediction dataset with default parameters",
        ),
    ]

    DEFAULT_CONFIG_NAME = "soybean-2018-10-5-std"

    def _create_builder_config(
        self,
        config_name=None,
        custom_features=None,
        **config_kwargs,
    ):
        """Create a BuilderConfig from config_kwargs.

        This method allows passing parameters directly to load_dataset() like:
        load_dataset("path", crop_type="corn", test_year=2017, ...)
        """
        # If config_name is provided and matches existing config, use it
        if config_name and config_name in [
            config.name for config in self.BUILDER_CONFIGS
        ]:
            for config in self.BUILDER_CONFIGS:
                if config.name == config_name:
                    return config, config_name

        # Otherwise, create a new config from the provided parameters
        if config_kwargs:
            # Create new config with provided parameters
            config = CropYieldConfig(**config_kwargs)
            return config, config.name

        # Fall back to default behavior
        return super()._create_builder_config(
            config_name=config_name,
            custom_features=custom_features,
            **config_kwargs,
        )

    def _info(self):
        # Get n_past_years from config, with fallback to default
        n_past_years = (
            self.config.n_past_years if self.config.n_past_years is not None else 5
        )

        # Calculate concrete shapes based on n_past_years
        seq_len = 52 * (n_past_years + 1)  # 52 weeks per year * number of years
        n_weather_vars = (
            5  # precipitation, reference_et, snow_lwe, solar_radiation, t2m_max
        )
        n_land_surface_vars = 3  # lai_high, lai_low, ndvi
        n_soil_vars = 8  # Different soil properties
        soil_depths = 6  # Different depth layers

        features = datasets.Features(
            {
                "weather": datasets.Array2D(
                    shape=(seq_len, n_weather_vars), dtype="float32"
                ),
                "land_surface": datasets.Array2D(
                    shape=(seq_len, n_land_surface_vars), dtype="float32"
                ),
                "land_surface_mask": datasets.Array2D(
                    shape=(seq_len, n_land_surface_vars), dtype="bool"
                ),
                "soil": datasets.Array2D(
                    shape=(soil_depths, n_soil_vars), dtype="float32"
                ),
                "coords": datasets.Array2D(shape=(1, 2), dtype="float32"),
                "years": datasets.Array2D(shape=(n_past_years + 1, 1), dtype="float32"),
                "y_past": datasets.Sequence(
                    feature=datasets.Value("float32"), length=n_past_years
                ),
                "y": datasets.Sequence(feature=datasets.Value("float32"), length=1),
            }
        )

        return datasets.DatasetInfo(
            description="Crop yield prediction dataset with weather and land surface data",
            features=features,
        )

    def _split_generators(self, dl_manager):
        # Get configuration parameters
        config = self.config
        if not isinstance(config, CropYieldConfig):
            # Fallback to default values if config is not CropYieldConfig
            logging.warning(
                "Config is not CropYieldConfig, using default values for soybean"
            )
            test_year = 2018
            n_train_years = 10
            n_past_years = 5
            data_dir = "./"
            standardize = True
            crop = "soybean"
        else:
            test_year = config.test_year
            n_train_years = config.n_train_years
            n_past_years = config.n_past_years
            data_dir = config.data_dir
            standardize = config.standardize
            crop = config.crop_type  # Use crop_type instead of name

        # Ensure data_dir is not None and has proper format
        if data_dir is None:
            data_dir = "./"
        elif not data_dir.endswith("/"):
            data_dir += "/"

        # Read the dataset
        crop_df = self._read_crop_dataset(data_dir, crop)

        return [
            datasets.SplitGenerator(
                name="train",
                gen_kwargs={
                    "data": crop_df,
                    "test_year": test_year,
                    "n_train_years": n_train_years,
                    "n_past_years": n_past_years,
                    "crop": crop,
                    "standardize": standardize,
                    "is_test": False,
                },
            ),
            datasets.SplitGenerator(
                name="test",
                gen_kwargs={
                    "data": crop_df,
                    "test_year": test_year,
                    "n_train_years": n_train_years,
                    "n_past_years": n_past_years,
                    "crop": crop,
                    "standardize": standardize,
                    "is_test": True,
                },
            ),
        ]

    def _generate_examples(
        self, data, test_year, n_train_years, n_past_years, crop, standardize, is_test
    ):
        # Process and standardize data
        processed_data = self._process_data(
            data, test_year, n_train_years, crop, standardize
        )

        # Create dataset samples
        dataset_samples = self._create_dataset_samples(
            processed_data, test_year, n_train_years, n_past_years, crop, is_test
        )

        for idx, sample in enumerate(dataset_samples):
            yield idx, sample

    def _read_crop_dataset(self, data_dir: str, crop: str):
        """Load and merge separate CSV files for Argentina data"""

        # Define file paths for Argentina processed CSVs
        csv_dir = os.path.join(data_dir, "data", "argentina", "processed", "csvs")

        # Load crop yield data
        crop_file = f"crop_{crop}_yield_1970-2024.csv"
        if crop == "wheat":
            crop_file = f"crop_{crop}_yield_1970-2025.csv"

        crop_path = os.path.join(csv_dir, crop_file)
        crop_df = pd.read_csv(crop_path)

        # Load weather data files
        weather_files = {
            "precipitation": "weather_1979-2024_precipitation_weekly_weighted_admin2.csv",
            "reference_et": "weather_1979-2024_reference_et_weekly_weighted_admin2.csv",
            "snow_lwe": "weather_1979-2024_snow_lwe_weekly_weighted_admin2.csv",
            "solar_radiation": "weather_1979-2024_solar_radiation_weekly_weighted_admin2.csv",
            "t2m_max": "weather_1979-2024_t2m_max_weekly_weighted_admin2.csv",
        }

        # Load land surface data files
        land_surface_files = {
            "lai_high": "land_surface_1979-2024_lai_high_weekly_weighted_admin2.csv",
            "lai_low": "land_surface_1979-2024_lai_low_weekly_weighted_admin2.csv",
            "ndvi": "land_surface_1982-2024_ndvi_weekly_weighted_admin2.csv",
        }

        # Load soil data files
        soil_files = {
            "cec": "soil_cec_weighted_admin2.csv",
            "coarse_fragments": "soil_coarse_fragments_weighted_admin2.csv",
            "nitrogen": "soil_nitrogen_weighted_admin2.csv",
            "organic_carbon": "soil_organic_carbon_weighted_admin2.csv",
            "organic_carbon_density": "soil_organic_carbon_density_weighted_admin2.csv",
            "ph_h2o": "soil_ph_h2o_weighted_admin2.csv",
            "sand": "soil_sand_weighted_admin2.csv",
            "silt": "soil_silt_weighted_admin2.csv",
        }

        # Load all weather and land surface data
        weather_dfs = {}
        for var_name, filename in weather_files.items():
            file_path = os.path.join(csv_dir, filename)
            df = pd.read_csv(file_path)
            weather_dfs[var_name] = df

        land_surface_dfs = {}
        for var_name, filename in land_surface_files.items():
            file_path = os.path.join(csv_dir, filename)
            df = pd.read_csv(file_path)
            land_surface_dfs[var_name] = df

        soil_dfs = {}
        for var_name, filename in soil_files.items():
            file_path = os.path.join(csv_dir, filename)
            if os.path.exists(file_path):
                df = pd.read_csv(file_path)
                soil_dfs[var_name] = df
            else:
                logging.warning(f"Soil file {filename} not found, skipping")

        # Create location identifier for merging
        crop_df["loc_ID"] = crop_df["admin_level_1"] + "_" + crop_df["admin_level_2"]

        # Start with crop data as base
        merged_df = crop_df.copy()

        # Add latitude and longitude from weather data (they should be consistent)
        if "precipitation" in weather_dfs:
            precip_df = weather_dfs["precipitation"]
            precip_df["loc_ID"] = (
                precip_df["admin_level_1"] + "_" + precip_df["admin_level_2"]
            )

            # Add lat/lng to merged_df
            lat_lng_df = precip_df[["loc_ID", "year", "latitude", "longitude"]].copy()
            merged_df = merged_df.merge(lat_lng_df, on=["loc_ID", "year"], how="left")
            merged_df = merged_df.rename(
                columns={"latitude": "lat", "longitude": "lng"}
            )

        # Merge weather data
        for var_name, df in weather_dfs.items():
            df["loc_ID"] = df["admin_level_1"] + "_" + df["admin_level_2"]

            # Get weekly columns for this variable
            week_cols = [
                col for col in df.columns if col.startswith(f"{var_name}_week_")
            ]

            # Rename columns to match expected format (W_varindex_weeknum)
            var_index = list(weather_files.keys()).index(var_name) + 1  # 1-indexed
            rename_dict = {}
            for i, col in enumerate(week_cols, 1):
                rename_dict[col] = f"W_{var_index}_{i}"

            df_renamed = df[["loc_ID", "year"] + week_cols].rename(columns=rename_dict)
            merged_df = merged_df.merge(df_renamed, on=["loc_ID", "year"], how="left")

        # Merge land surface data
        for var_name, df in land_surface_dfs.items():
            df["loc_ID"] = df["admin_level_1"] + "_" + df["admin_level_2"]

            # Get weekly columns for this variable
            week_cols = [
                col for col in df.columns if col.startswith(f"{var_name}_week_")
            ]

            # Continue indexing from where weather variables left off
            var_index = (
                len(weather_files) + list(land_surface_files.keys()).index(var_name) + 1
            )
            rename_dict = {}
            for i, col in enumerate(week_cols, 1):
                rename_dict[col] = f"W_{var_index}_{i}"

            df_renamed = df[["loc_ID", "year"] + week_cols].rename(columns=rename_dict)
            merged_df = merged_df.merge(df_renamed, on=["loc_ID", "year"], how="left")

        # Merge soil data (assuming soil data has depth columns)
        for var_name, df in soil_dfs.items():
            df["loc_ID"] = df["admin_level_1"] + "_" + df["admin_level_2"]

            # Get depth columns for this variable (assuming format like cec_0_5cm, cec_5_15cm, etc.)
            depth_cols = [
                col
                for col in df.columns
                if col.startswith(f"{var_name}_") and "cm" in col
            ]

            if depth_cols:
                # Continue indexing from where land surface variables left off
                var_index = (
                    len(weather_files)
                    + len(land_surface_files)
                    + list(soil_files.keys()).index(var_name)
                    + 1
                )
                rename_dict = {}
                for i, col in enumerate(depth_cols, 1):
                    rename_dict[col] = f"S_{var_index}_{i}"

                df_renamed = df[["loc_ID"] + depth_cols].rename(columns=rename_dict)
                # For soil data, merge only on loc_ID (soil properties don't change by year)
                merged_df = merged_df.merge(df_renamed, on=["loc_ID"], how="left")

        # Sort by location and year
        merged_df = merged_df.sort_values(["loc_ID", "year"])

        logging.info(f"Loaded {len(merged_df)} records for {crop} from Argentina data")
        logging.info(
            f"Data covers years {merged_df['year'].min()}-{merged_df['year'].max()}"
        )
        logging.info(f"Number of unique locations: {merged_df['loc_ID'].nunique()}")

        return merged_df

    def _process_data(self, data, test_year, n_train_years, crop, standardize):
        start_year = test_year - n_train_years

        data = data[data["year"] > 1981.0].copy()

        # Drop rows with missing yield values for the given crop
        yield_col = f"{crop}_yield"
        rows_before = len(data)
        data = data.dropna(subset=[yield_col])
        rows_after = len(data)
        rows_dropped = rows_before - rows_after

        if rows_dropped > 0:
            print(
                f"Dropped {rows_dropped} rows with missing {yield_col} values ({rows_before} -> {rows_after} rows)"
            )

        if standardize:
            # Standardize data
            cols_to_standardize = [
                col
                for col in data.columns
                if col
                not in [
                    "loc_ID",
                    "year",
                    "country",
                    "admin_level_1",
                    "admin_level_2",
                    "lat",
                    "lng",
                    yield_col,
                ]
            ]

            data[cols_to_standardize] = (
                data[cols_to_standardize] - data[cols_to_standardize].mean()
            ) / data[cols_to_standardize].std()
            data[cols_to_standardize] = data[cols_to_standardize].fillna(0)

            # Standardize yield data
            train_data = data[(data["year"] >= start_year) & (data["year"] < test_year)]
            yield_mean, yield_std = (
                train_data[yield_col].mean(),
                train_data[yield_col].std(),
            )
            data[yield_col] = (data[yield_col] - yield_mean) / yield_std

            print(f"{crop} yield mean = {yield_mean:.3f} and std = {yield_std:.3f}")
            CROP_YIELD_STATS[crop]["mean"].append(yield_mean)
            CROP_YIELD_STATS[crop]["std"].append(yield_std)
        # If not standardizing, still need to handle NaN values
        data = data.fillna(0)

        return data

    def _create_dataset_samples(
        self, data, test_year, n_train_years, n_past_years, crop, is_test
    ):
        start_year = test_year - n_train_years
        yield_col = f"{crop}_yield"

        # Define column groups for Argentina data
        weather_cols = [
            f"W_{i}_{j}" for i in range(1, 6) for j in range(1, 53)
        ]  # 5 weather variables, 52 weeks

        land_surface_cols = [
            f"W_{i}_{j}" for i in range(6, 9) for j in range(1, 53)
        ]  # 3 land surface variables, 52 weeks

        # Check if soil data exists in the dataset
        soil_cols = [
            f"S_{i}_{j}" for i in range(1, 9) for j in range(1, 7)
        ]  # 8 soil variables, 6 depth layers

        # Filter candidate data
        if is_test:
            candidate_data = data[data["year"] == test_year]
        else:
            candidate_data = data[
                (data["year"] >= start_year) & (data["year"] < test_year)
            ]

        # Filter to only include cases where we have complete historical data
        data_sorted = data.sort_values(["loc_ID", "year"])

        def has_sufficient_history(row):
            year, loc_ID = row["year"], row["loc_ID"]
            loc_data = data_sorted[data_sorted["loc_ID"] == loc_ID]
            loc_data_up_to_year = loc_data[loc_data["year"] <= year]
            return len(loc_data_up_to_year.tail(n_past_years + 1)) == n_past_years + 1

        mask = candidate_data.apply(has_sufficient_history, axis=1)
        valid_candidates = candidate_data[mask]
        index = valid_candidates[["year", "loc_ID"]].reset_index(drop=True)

        dataset_name = "train" if not is_test else "test"
        logging.info(
            f"Creating {dataset_name} dataset with {len(index)} samples for {'test year ' + str(test_year) if is_test else 'training years ' + str(start_year) + '-' + str(test_year-1)} using {crop} yield."
        )

        samples = []
        total_samples = len(index)

        if total_samples == 0:
            logging.warning(f"No samples found for {dataset_name} dataset!")
            return samples

        for idx in tqdm(range(total_samples)):
            year = int(index.iloc[idx]["year"])
            loc_ID = index.iloc[idx]["loc_ID"]
            query_data = data[(data["year"] <= year) & (data["loc_ID"] == loc_ID)].tail(
                n_past_years + 1
            )

            # Extract weather data
            weather_data = (
                query_data[weather_cols]
                .values.astype("float32")
                .reshape((-1, N_WEATHER_VARS, 52))
            )
            n_years, n_weather_features, seq_len = weather_data.shape
            total_seq_len = n_years * seq_len

            if total_seq_len > MAX_CONTEXT_LENGTH:
                raise ValueError(
                    f"total_seq_len = {total_seq_len} is greater than MAX_CONTEXT_LENGTH = {MAX_CONTEXT_LENGTH}"
                )

            # Reshape weather: (n_years, n_features, 52) -> (total_seq_len, n_features)
            weather = weather_data.transpose(0, 2, 1).reshape(
                total_seq_len, n_weather_features
            )

            # Extract land surface data
            land_surface_data = (
                query_data[land_surface_cols]
                .values.astype("float32")
                .reshape((-1, N_LAND_SURFACE_VARS, 52))
            )
            # Reshape land surface: (n_years, n_features, 52) -> (total_seq_len, n_features)
            land_surface = land_surface_data.transpose(0, 2, 1).reshape(
                total_seq_len, N_LAND_SURFACE_VARS
            )

            # Create land surface mask (True where NDVI data is missing before 1982)
            land_surface_mask = np.zeros(
                (total_seq_len, N_LAND_SURFACE_VARS), dtype=bool
            )
            year_data = query_data["year"].values
            for i, year_val in enumerate(year_data):
                start_idx = i * seq_len
                end_idx = (i + 1) * seq_len
                if year_val < 1982:  # NDVI starts from 1982
                    land_surface_mask[start_idx:end_idx, 2] = (
                        True  # Mask NDVI (index 2)
                    )

            # Extract soil data (static for each location)
            soil_data = (
                query_data[soil_cols]
                .iloc[0]
                .values.astype("float32")
                .reshape((N_SOIL_VARS, SOIL_DEPTHS))
            )
            soil = soil_data.T  # Transpose to get (soil_depths, n_soil_vars)

            # Extract coordinates (single location)
            coords = (
                query_data[["lat", "lng"]]
                .iloc[0]
                .values.astype("float32")
                .reshape((1, 2))
            )

            # Extract years (n_past_years + 1, 1)
            years = query_data["year"].values.astype("float32").reshape((-1, 1))

            # Extract yield data
            y = query_data.iloc[-1:][yield_col].values.astype("float32")
            y_past = query_data[yield_col].values.astype("float32")[
                :-1
            ]  # Exclude current year

            if len(y_past) < n_past_years:
                raise ValueError(
                    f"Insufficient yield history for location {loc_ID} in year {year}. Need {n_past_years} past years but have {len(y_past)}."
                )

            sample = {
                "weather": weather,
                "land_surface": land_surface,
                "land_surface_mask": land_surface_mask,
                "soil": soil,
                "coords": coords,
                "years": years,
                "y_past": y_past,
                "y": y,
            }

            samples.append(sample)

        return samples