weather_forecast_discussion/dataset.py

import os
import pandas as pd
import numpy as np
import datasets

class CSVhrrrDataset(datasets.GeneratorBasedBuilder):
    """
    A custom dataset to load CSV and corresponding hrrr files.
    The CSV files are loaded using pandas and the hrrr files using numpy.
    """

    # Define dataset name and version
    VERSION = datasets.Version("1.0.0")

    def _info(self):
        # Dataset description and features
        return datasets.DatasetInfo(
            description="Dataset containing CSV and corresponding hrrr files.",
            features=datasets.Features({
                "csv_data": datasets.Value("string"),  # CSV content as a string or specific columns
                "hrrr_file_path": datasets.Value("string"),  # Assuming hrrr files contain float32 data
                "filename": datasets.Value("string"),  # Filename of the CSV file
            }),
            supervised_keys=None,
            homepage="https://huggingface.co/datasets/nasa-impact/WINDSET/tree/main/weather_forecast_discussion",
            license="MIT",
        )

    def _split_generators(self, dl_manager):
        """
        Define dataset splits for this dataset (train, validation, test).
        In this case, we just load all the files into a single split.
        """
        # Get the directory paths
        csv_dir = os.path.join(os.getcwd(), "weather_forecast_discussion/csv_reports")
        hrrr_dir = os.path.join(os.getcwd(), "weather_forecast_discussion/hrrr")

        # Check that both directories exist
        if not os.path.isdir(csv_dir):
            raise FileNotFoundError(f"CSV directory {csv_dir} not found!")
        if not os.path.isdir(hrrr_dir):
            raise FileNotFoundError(f"hrrr directory {hrrr_dir} not found!")

        # List CSV and hrrr files
        csv_files = [f for f in os.listdir(csv_dir) if f.endswith('.csv')]
        hrrr_files = [f for f in os.listdir(hrrr_dir) if f.endswith('.grib2')]

        # Ensure CSV and hrrr files are paired correctly by date
        file_pairs = []
        for csv_file in csv_files:
            # Extract the date from the CSV file (assuming format is 'date.csv')
            date_str = os.path.splitext(csv_file)[0]
            
            # Search for matching hrrr files in the hrrr directory
            matching_hrrr_files = [hrrr for hrrr in hrrr_files if f"hrrr.{date_str}." in hrrr]

            if len(matching_hrrr_files) == 1:  # Ensure exactly one match
                file_pairs.append((csv_file, matching_hrrr_files[0]))
            elif len(matching_hrrr_files) == 0:
                print(f"Warning: No matching hrrr file found for CSV file: {csv_file}")
            else:
                print(f"Warning: Multiple matching hrrr files found for CSV file: {csv_file}. Using the first match.")
                file_pairs.append((csv_file, matching_hrrr_files[0]))  # Use the first match if multiple are found

        # If no valid file pairs were found, raise an error
        if not file_pairs:
            raise ValueError("No valid CSV-hrrr file pairs found. Check the directory structure and file names.")

        # Create a split generator
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "file_pairs": file_pairs,
                    "csv_dir": csv_dir,
                    "hrrr_dir": hrrr_dir,
                }
            )
        ]

    def _generate_examples(self, file_pairs, csv_dir, hrrr_dir):
        """
        Yield examples from the CSV and hrrr files.
        Each example contains data from a CSV file and its corresponding hrrr file.
        """
        example_id = 0

        for csv_file, hrrr_file in file_pairs:
            # Load CSV file using pandas
            csv_file_path = os.path.join(csv_dir, csv_file)
            csv_data = pd.read_csv(csv_file_path)

            # Load corresponding hrrr file using numpy
            hrrr_file_path = os.path.join(hrrr_dir, hrrr_file)

            # Yield example with both CSV and hrrr data
            yield example_id, {
                "csv_data": csv_data["discussion"].to_string(),  # Store content under discussion only
                "hrrr_file_path": hrrr_file_path,  
                "filename": csv_file, 
            }
            example_id += 1