automatic download

substrate_mix_raman.py

@@ -0,0 +1,107 @@
+import datasets
+import pandas as pd
+import os
+
+_DESCRIPTION = """
+This dataset contains Raman spectra of various mixtures of 8 biotechnologically relevant substances, along with concentrations of mineral salt medium and antifoam.
+The dataset is designed for calibration resources in high-throughput bioprocess development, featuring statistically independent substance concentrations.
+"""
+
+_HOMEPAGE = "https://huggingface.co/datasets/chlange/SubstrateMixRaman" # Replace if you have an official project page
+_LICENSE = "https://creativecommons.org/licenses/by/4.0/" # Make sure this matches your README.md license
+
+# _URL is no longer needed because data_files in README.md handles split paths
+
+class SubstrateMixRaman(datasets.GeneratorBasedBuilder):
+    VERSION = datasets.Version("1.0.0")
+
+    def _info(self):
+        # Define the features of your dataset
+        # The first 1561 columns are Raman intensities (spectra)
+        # We don't need to name them individually here for the 'spectrum' feature,
+        # as it's a sequence.
+        # If you wanted the actual Raman shift values as a feature, you'd add:
+        # "raman_shifts": datasets.Sequence(datasets.Value("float32")),
+
+        # The next 8 columns are concentrations (labels)
+        label_features = {
+            "Glucose": datasets.Value("float32"),
+            "Glycerol": datasets.Value("float32"),
+            "Acetate": datasets.Value("float32"),
+            "EnPump": datasets.Value("float32"),
+            "Nitrate": datasets.Value("float32"),
+            "Yeast_Extract": datasets.Value("float32"),
+            "total_phosphate": datasets.Value("float32"), # Renamed as per CSV header for clarity
+            "total_sulfate": datasets.Value("float32"),   # Renamed as per CSV header for clarity
+        }
+
+        features = datasets.Features({
+            "spectrum": datasets.Sequence(datasets.Value("float32")), # A list of floats for the spectrum
+            "labels": datasets.Features(label_features) # Nested features for concentrations
+        })
+
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=features,
+            homepage=_HOMEPAGE,
+            license=_LICENSE,
+            citation="""
+            @article{lange5239248deep,
+            title={Deep Learning for Raman Spectroscopy: Benchmarking Models for Upstream Bioprocess Monitoring},
+            author={Lange, Christoph and Altmann, Madeline and Stors, Daniel and Seidel, Simon and Moynahan, Kyle and Cai, Linda and Born, Stefan and Neubauer, Peter and Cruz Bournazou, Nicolas},
+            journal={Available at SSRN 5239248}
+            }
+            """ 
+        )
+
+    def _split_generators(self, dl_manager):
+        """Returns SplitGenerators."""
+        # dl_manager.download handles the data_files specification from README.md
+        # self.config.data_files will automatically be populated based on the YAML
+        downloaded_files = dl_manager.download_and_extract(self.config.data_files)
+
+        splits = []
+        for split_name, filepath in downloaded_files.items():
+            splits.append(
+                datasets.SplitGenerator(
+                    name=split_name, # This will be 'train', 'validation', 'test'
+                    gen_kwargs={"filepath": filepath},
+                )
+            )
+        return splits
+
+    def _generate_examples(self, filepath):
+        """Yields examples as (key, example) tuples for a given split file."""
+        df = pd.read_csv(filepath)
+
+        # Assuming the actual column names in your CSV are the Raman shifts as strings
+        # and then the specified label names.
+        # Let's verify the number of spectrum columns based on your description.
+        # 1561 columns for spectrum and 8 for labels = 1569 columns total.
+        num_spectrum_cols = 1561
+        raman_shift_columns = [str(col) for col in df.columns[:num_spectrum_cols]]
+
+        label_columns = [
+            "Glucose", "Glycerol", "Acetate", "EnPump", "Nitrate",
+            "Yeast_Extract", "total_phosphate", "total_sulfate"
+        ]
+
+        # Ensure all expected label columns are actually present in the DataFrame
+        # This can help debug if there's a mismatch between the script and the CSV.
+        for col in label_columns:
+            if col not in df.columns:
+                raise ValueError(f"Label column '{col}' not found in CSV file: {filepath}")
+
+        for id_, row in enumerate(df.iterrows()):
+            data = row[1] # row[0] is the index, row[1] is the Series/array of data
+
+            # Extract spectrum (Raman intensities)
+            spectrum = [float(data[col]) for col in raman_shift_columns]
+
+            # Extract labels (concentrations)
+            labels = {label_col: float(data[label_col]) for label_col in label_columns}
+
+            yield id_, {
+                "spectrum": spectrum,
+                "labels": labels,
+            }