Update BubbleML_2.py

BubbleML_2.py

@@ -1,98 +1,132 @@
+# BubbleML_2.py
+
 import os
-import h5py
 import json
+import h5py
 import numpy as np
+
 import datasets
+from datasets import (
+    BuilderConfig,
+    GeneratorBasedBuilder,
+    DatasetInfo,
+    Features,
+    Array3D,
+    Sequence,
+    Value,
+    Split,
+    Version,
+)
 
+_CITATION = ""  # optional
+_DESCRIPTION = """
+BubbleML: high-fidelity single-bubble boiling simulations (FC72 & R515B).
+Pre-defined train/test splits across two directories.
+"""
 
-class BubbleMLConfig(datasets.BuilderConfig):
-    def __init__(self, subset_dirs, train_files_dict, test_files_dict, **kwargs):
-        super().__init__(version=datasets.Version("1.0.0"), **kwargs)
-        self.subset_dirs = subset_dirs  # e.g., ["SingleBubble-Saturated-FC72-2D", "SingleBubble-Saturated-R515B-2D"]
-        self.train_files_dict = train_files_dict  # dict mapping dir -> list of filenames
+class BubbleMLConfig(BuilderConfig):
+    """BuilderConfig for BubbleML_2."""
+    def __init__(
+        self,
+        subset_dirs,
+        train_files_dict,
+        test_files_dict,
+        version: Version = Version("1.0.0"),
+        **kwargs,
+    ):
+        super().__init__(name=kwargs.pop("name"), version=version, description=kwargs.pop("description"), **kwargs)
+        self.subset_dirs = subset_dirs
+        self.train_files_dict = train_files_dict
         self.test_files_dict = test_files_dict
 
-
-class BubbleMLDataset(datasets.GeneratorBasedBuilder):
+class BubbleMLDataset(GeneratorBasedBuilder):
+    """BubbleML_2: combined single-bubble dataset."""
+    
+    BUILDER_CONFIG_CLASS = BubbleMLConfig
     BUILDER_CONFIGS = [
         BubbleMLConfig(
             name="single-bubble",
-            description="Combined dataset for single bubble boiling (FC72 and R515B)",
-            subset_dirs=["SingleBubble-Saturated-FC72-2D", "SingleBubble-Saturated-R515B-2D"],
+            description="Single-bubble (FC72 & R515B) train/test split",
+            subset_dirs=[
+                "SingleBubble-Saturated-FC72-2D",
+                "SingleBubble-Saturated-R515B-2D",
+            ],
             train_files_dict={
                 "SingleBubble-Saturated-FC72-2D": [
-                    "Twall_90.hdf5", "Twall_91.hdf5", "Twall_92.hdf5", "Twall_94.hdf5",
-                    "Twall_96.hdf5", "Twall_98.hdf5", "Twall_99.hdf5", "Twall_100.hdf5"
+                    "Twall_90.hdf5","Twall_91.hdf5","Twall_92.hdf5","Twall_94.hdf5",
+                    "Twall_96.hdf5","Twall_98.hdf5","Twall_99.hdf5","Twall_100.hdf5",
                 ],
                 "SingleBubble-Saturated-R515B-2D": [
-                    "Twall_13.hdf5", "Twall_14.hdf5", "Twall_15.hdf5", "Twall_17.hdf5",
-                    "Twall_19.hdf5", "Twall_21.hdf5", "Twall_22.hdf5", "Twall_23.hdf5"
+                    "Twall_13.hdf5","Twall_14.hdf5","Twall_15.hdf5","Twall_17.hdf5",
+                    "Twall_19.hdf5","Twall_21.hdf5","Twall_22.hdf5","Twall_23.hdf5",
                 ],
             },
             test_files_dict={
-                "SingleBubble-Saturated-FC72-2D": ["Twall_87.hdf5", "Twall_95.hdf5", "Twall_103.hdf5"],
-                "SingleBubble-Saturated-R515B-2D": ["Twall_10.hdf5", "Twall_18.hdf5", "Twall_26.hdf5"],
-            }
-        )
+                "SingleBubble-Saturated-FC72-2D": ["Twall_87.hdf5","Twall_95.hdf5","Twall_103.hdf5"],
+                "SingleBubble-Saturated-R515B-2D": ["Twall_10.hdf5","Twall_18.hdf5","Twall_26.hdf5"],
+            },
+        ),
     ]
+    DEFAULT_CONFIG_NAME = "single-bubble"
 
-    def _info(self):
-        return datasets.DatasetInfo(
-            description="BubbleML: Boiling simulations with predefined train/test splits used in BubbleFormer.",
-            features=datasets.Features({
-                "input": datasets.Array3D(shape=(5, 4, None, None), dtype="float32"),
-                "output": datasets.Array3D(shape=(5, 4, None, None), dtype="float32"),
-                "fluid_params": datasets.Sequence(datasets.Value("float32"))
+    def _info(self) -> DatasetInfo:
+        return DatasetInfo(
+            description=_DESCRIPTION,
+            features=Features({
+                "input": Array3D(shape=(5, 4, None, None), dtype="float32"),
+                "output": Array3D(shape=(5, 4, None, None), dtype="float32"),
+                "fluid_params": Sequence(Value("float32")),
+                "filename": Value("string"),
             }),
             supervised_keys=None,
+            homepage="https://huggingface.co/datasets/hpcforge/BubbleML_2",
+            citation=_CITATION,
         )
 
-    def _split_generators(self, dl_manager):
-        config = self.config
-        base_dir = dl_manager.download_and_extract(".")
+    def _split_generators(self, dl_manager: datasets.DownloadManager):
+        # data already lives in the repo
+        base_dir = dl_manager.extract("./")
+        cfg = self.config
 
-        def resolve_paths(files_dict):
-            resolved = []
-            for subdir, files in files_dict.items():
-                for f in files:
-                    resolved.append(os.path.join(base_dir, subdir, f))
-            return resolved
-
-        train_files = resolve_paths(config.train_files_dict)
-        test_files = resolve_paths(config.test_files_dict)
+        def _resolve(files_dict):
+            paths = []
+            for d, files in files_dict.items():
+                for fname in files:
+                    paths.append(os.path.join(base_dir, d, fname))
+            return paths
 
         return [
-            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"files": train_files}),
-            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"files": test_files}),
+            datasets.SplitGenerator(
+                name=Split.TRAIN,
+                gen_kwargs={"files": _resolve(cfg.train_files_dict)},
+            ),
+            datasets.SplitGenerator(
+                name=Split.TEST,
+                gen_kwargs={"files": _resolve(cfg.test_files_dict)},
+            ),
         ]
 
     def _generate_examples(self, files):
-        for idx, fpath in enumerate(files):
-            with h5py.File(fpath, "r") as h5f:
-                input_data = np.stack([h5f[k][:5] for k in ["dfun", "temperature", "velx", "vely"]])
-                output_data = np.stack([h5f[k][5:10] for k in ["dfun", "temperature", "velx", "vely"]])
-
-            # Read matching JSON metadata
-            json_path = fpath.replace(".hdf5", ".json")
-            if os.path.exists(json_path):
-                with open(json_path, "r") as jf:
-                    params = json.load(jf)
+        for idx, path in enumerate(files):
+            with h5py.File(path, "r") as h5f:
+                inp = np.stack([h5f[k][:5] for k in ["dfun","temperature","velx","vely"]])
+                out = np.stack([h5f[k][5:10] for k in ["dfun","temperature","velx","vely"]])
+            # metadata
+            meta_path = path.replace(".hdf5", ".json")
+            if os.path.exists(meta_path):
+                with open(meta_path) as jf:
+                    p = json.load(jf)
                 fluid_params = [
-                    params["inv_reynolds"],
-                    params["cpgas"],
-                    params["mugas"],
-                    params["rhogas"],
-                    params["thcogas"],
-                    params["stefan"],
-                    params["prandtl"],
-                    params["heater"]["nucWaitTime"],
-                    params["heater"]["wallTemp"],
+                    p["inv_reynolds"], p["cpgas"], p["mugas"], p["rhogas"],
+                    p["thcogas"], p["stefan"], p["prandtl"],
+                    p["heater"]["nucWaitTime"], p["heater"]["wallTemp"]
                 ]
             else:
                 fluid_params = []
 
             yield idx, {
-                "input": input_data.astype(np.float32),
-                "output": output_data.astype(np.float32),
-                "fluid_params": fluid_params
+                "input": inp.astype(np.float32),
+                "output": out.astype(np.float32),
+                "fluid_params": fluid_params,
+                "filename": os.path.basename(path),
             }