gengeneeval 0.2.0__py3-none-any.whl → 0.3.0__py3-none-any.whl

{gengeneeval-0.2.0.dist-info → gengeneeval-0.3.0.dist-info}/METADATA

@@ -1,10 +1,10 @@
 Metadata-Version: 2.4
 Name: gengeneeval
-Version: 0.2.0
-Summary: Comprehensive evaluation of generated gene expression data. Computes metrics between real and generated datasets with support for condition matching, train/test splits, and publication-quality visualizations.
+Version: 0.3.0
+Summary: Comprehensive evaluation of generated gene expression data. Computes metrics between real and generated datasets with support for condition matching, train/test splits, memory-efficient lazy loading, CPU parallelization, GPU acceleration, and publication-quality visualizations.
 License: MIT
 License-File: LICENSE
-Keywords: gene expression,evaluation,metrics,single-cell,generative models,benchmarking
+Keywords: gene expression,evaluation,metrics,single-cell,generative models,benchmarking,memory-efficient
 Author: GenEval Team
 Author-email: geneval@example.com
 Requires-Python: >=3.8,<4.0
@@ -24,6 +24,7 @@ Provides-Extra: full
 Provides-Extra: gpu
 Requires-Dist: anndata (>=0.8.0)
 Requires-Dist: geomloss (>=0.2.1) ; extra == "full" or extra == "gpu"
+Requires-Dist: joblib (>=1.0.0)
 Requires-Dist: matplotlib (>=3.5.0)
 Requires-Dist: numpy (>=1.21.0)
 Requires-Dist: pandas (>=1.3.0)
@@ -46,7 +47,7 @@ Description-Content-Type: text/markdown
 
 **Comprehensive evaluation of generated gene expression data against real datasets.**
 
-GenEval is a modular, object-oriented Python framework for computing metrics between real and generated gene expression datasets stored in AnnData (h5ad) format. It supports condition-based matching, train/test splits, and generates publication-quality visualizations.
+GenEval is a modular, object-oriented Python framework for computing metrics between real and generated gene expression datasets stored in AnnData (h5ad) format. It supports condition-based matching, train/test splits, memory-efficient lazy loading for large datasets, and generates publication-quality visualizations.
 
 ## Features
 
@@ -77,6 +78,10 @@ All metrics are computed **per-gene** (returning a vector) and **aggregated**:
 - ✅ Condition-based matching (perturbation, cell type, etc.)
 - ✅ Train/test split support
 - ✅ Per-gene and aggregate metrics
+- ✅ **Memory-efficient lazy loading** for large datasets
+- ✅ **Batched evaluation** to avoid OOM errors
+- ✅ **CPU parallelization** via joblib (multi-core speedup)
+- ✅ **GPU acceleration** via PyTorch (10-100x speedup)
 - ✅ Modular, extensible architecture
 - ✅ Command-line interface
 - ✅ Publication-quality visualizations
@@ -140,6 +145,108 @@ geneval --real real.h5ad --generated generated.h5ad \
         --output results/
 ```
 
+### Memory-Efficient Mode (for Large Datasets)
+
+For datasets too large to fit in memory, use the lazy evaluation API:
+
+```python
+from geneval import evaluate_lazy, load_data_lazy
+
+# Memory-efficient evaluation (streams data one condition at a time)
+results = evaluate_lazy(
+    real_path="large_real.h5ad",
+    generated_path="large_generated.h5ad",
+    condition_columns=["perturbation"],
+    batch_size=256,                    # Process in batches
+    use_backed=True,                   # Memory-mapped file access
+    output_dir="eval_output/",
+    save_per_condition=True,           # Save each condition to disk
+)
+
+# Get summary statistics
+print(results.get_summary())
+
+# Or use the lazy loader directly for custom workflows
+with load_data_lazy("real.h5ad", "gen.h5ad", ["perturbation"]) as loader:
+    print(f"Memory estimate: {loader.estimate_memory_usage()}")
+    
+    # Process one condition at a time
+    for key, real, gen, info in loader.iterate_conditions():
+        # Your custom evaluation logic
+        pass
+```
+
+### Accelerated Evaluation (CPU Parallelization & GPU)
+
+GenEval supports CPU parallelization and GPU acceleration for significant speedups:
+
+```python
+from geneval import evaluate, get_available_backends
+
+# Check available backends
+print(get_available_backends())
+# {'joblib': True, 'torch': True, 'geomloss': True, 'cuda': True, 'mps': False}
+
+# Parallel CPU evaluation (use all cores)
+results = evaluate(
+    real_path="real.h5ad",
+    generated_path="generated.h5ad",
+    condition_columns=["perturbation"],
+    n_jobs=-1,  # Use all available CPU cores
+)
+
+# GPU-accelerated evaluation
+results = evaluate(
+    real_path="real.h5ad",
+    generated_path="generated.h5ad",
+    condition_columns=["perturbation"],
+    device="cuda",  # Use NVIDIA GPU
+)
+
+# Combined: parallel CPU + auto device selection
+results = evaluate(..., n_jobs=8, device="auto")
+```
+
+#### Low-level Accelerated API
+
+For custom workflows, use the accelerated metrics directly:
+
+```python
+from geneval.metrics.accelerated import (
+    compute_metrics_accelerated,
+    GPUWasserstein1,
+    GPUMMD,
+    vectorized_wasserstein1,
+)
+import numpy as np
+
+# Load your data
+real = np.random.randn(1000, 5000)      # 1000 cells, 5000 genes
+generated = np.random.randn(1000, 5000)
+
+# Compute multiple metrics with acceleration
+results = compute_metrics_accelerated(
+    real, generated,
+    metrics=["wasserstein_1", "wasserstein_2", "mmd", "energy"],
+    n_jobs=8,          # CPU parallelization
+    device="cuda",     # GPU acceleration
+    verbose=True,
+)
+
+# Access results
+print(f"W1: {results['wasserstein_1'].aggregate_value:.4f}")
+print(f"MMD: {results['mmd'].aggregate_value:.4f}")
+```
+
+#### Performance Tips
+
+| Optimization | Speedup | When to Use |
+|--------------|---------|-------------|
+| `n_jobs=-1` (all cores) | 4-16x | Always (if joblib available) |
+| `device="cuda"` | 10-100x | Large datasets, NVIDIA GPU available |
+| `device="mps"` | 5-20x | Apple Silicon Macs |
+| Vectorized NumPy | 2-5x | Automatic fallback |
+
 ## Expected Data Format
 
 GenEval expects AnnData (h5ad) files with:

{gengeneeval-0.2.0.dist-info → gengeneeval-0.3.0.dist-info}/RECORD

@@ -1,15 +1,18 @@
-geneval/__init__.py,sha256=WB9yj2OLpY3tjw545G0xzSP8iP_fn4vOIWA2WaraEWk,2977
+geneval/__init__.py,sha256=K0E3Jyt3l7_KxqIeI3upBBBrjRA4ASdRFugaxMVVGRM,4306
 geneval/cli.py,sha256=0ai0IGyn3SSmEnfLRJhcr0brvUxuNZHE4IXod7jvosU,9977
 geneval/config.py,sha256=gkCjs_gzPWgUZNcmSR3Y70XQCAZ1m9AKLueaM-x8bvw,3729
 geneval/core.py,sha256=No0DP8bNR6LedfCWEedY9C5r_c4M14rvSPaGZqbxc94,1155
-geneval/data/__init__.py,sha256=nD3uWostZbYD3Yj_TOE44LvPDen-Vm3gN8ZH0QptPGw,450
+geneval/data/__init__.py,sha256=NQUPVpUnBIabrTH5TuRk0KE9S7sVO5QetZv-MCQmZuw,827
 geneval/data/gene_expression_datamodule.py,sha256=XiBIdf68JZ-3S-FaZsrQlBJA7qL9uUXo2C8y0r4an5M,8009
+geneval/data/lazy_loader.py,sha256=5fTRVjPjcWvYXV-uPWFUF2Nn9rHRdD8lygAUkCW8wOM,20677
 geneval/data/loader.py,sha256=zpRmwGZ4PJkB3rpXXRCMFtvMi4qvUrPkKmvIlGjfRpY,14555
-geneval/evaluator.py,sha256=wZFzLo2PLHanjA-9L6C3xJBjMWXxPM63kU6usU4P7bs,11619
+geneval/evaluator.py,sha256=WgdrgqOcGYT35k1keiFEIIRIj2CQaD2DsmBpq9hcLrI,13440
 geneval/evaluators/__init__.py,sha256=i11sHvhsjEAeI3Aw9zFTPmCYuqkGxzTHggAKehe3HQ0,160
 geneval/evaluators/base_evaluator.py,sha256=yJL568HdNofIcHgNOElSQMVlG9oRPTTDIZ7CmKccRqs,5967
 geneval/evaluators/gene_expression_evaluator.py,sha256=v8QL6tzOQ3QVXdPMM8tFHTTviZC3WsPRX4G0ShgeDUw,8743
-geneval/metrics/__init__.py,sha256=H5IXTKR-zoP_pGht6ioJfhLU7IHrSDQElMk0Cp4-JTw,1786
+geneval/lazy_evaluator.py,sha256=I_VvDolxPFGiW38eGPrjSoBOKICKyYN3GHbjJBAe5tg,13200
+geneval/metrics/__init__.py,sha256=yVlNcFxfudOE4q-Y1VNJIXw1HrM70LkxocJgg3Cp7vo,2359
+geneval/metrics/accelerated.py,sha256=iVxXg1Bf4aAeh-0kz7JRZS1I7xHHy9vNRozGDmCY_QY,27364
 geneval/metrics/base_metric.py,sha256=prbnB-Ap-P64m-2_TUrHxO3NFQaw-obVg1Tw4pjC5EY,6961
 geneval/metrics/correlation.py,sha256=jpYmaihWK89J1E5yQinGUJeB6pTZ21xPNHJi3XYyXJE,6987
 geneval/metrics/distances.py,sha256=9mWzbMbIBY1ckOd2a0l3by3aEFMQZL9bVMSeP44xzUg,16155
@@ -25,8 +28,8 @@ geneval/utils/preprocessing.py,sha256=1Cij1O2dwDR6_zh5IEgLPq3jEmV8VfIRjfQrHiKe3M
 geneval/visualization/__init__.py,sha256=LN19jl5xV4WVJTePaOUHWvKZ_pgDFp1chhcklGkNtm8,792
 geneval/visualization/plots.py,sha256=3K94r3x5NjIUZ-hYVQIivO63VkLOvDWl-BLB_qL2pSY,15008
 geneval/visualization/visualizer.py,sha256=lX7K0j20nAsgdtOOdbxLdLKYAfovEp3hNAnZOjFTCq0,36670
-gengeneeval-0.2.0.dist-info/METADATA,sha256=LyE9iQQUOMakv91vebWNBcg10quJatQwYHGucqri3Rw,6262
-gengeneeval-0.2.0.dist-info/WHEEL,sha256=zp0Cn7JsFoX2ATtOhtaFYIiE2rmFAD4OcMhtUki8W3U,88
-gengeneeval-0.2.0.dist-info/entry_points.txt,sha256=xTkwnNa2fP0w1uGVsafzRTaCeuBSWLlNO-1CN8uBSK0,43
-gengeneeval-0.2.0.dist-info/licenses/LICENSE,sha256=RDHgHDI4rSDq35R4CAC3npy86YUnmZ81ecO7aHfmmGA,1073
-gengeneeval-0.2.0.dist-info/RECORD,,
+gengeneeval-0.3.0.dist-info/METADATA,sha256=5K2bIh59OEM88dNVeUWPOevyyAbnAIyiKaZu6VmJIh0,9680
+gengeneeval-0.3.0.dist-info/WHEEL,sha256=zp0Cn7JsFoX2ATtOhtaFYIiE2rmFAD4OcMhtUki8W3U,88
+gengeneeval-0.3.0.dist-info/entry_points.txt,sha256=xTkwnNa2fP0w1uGVsafzRTaCeuBSWLlNO-1CN8uBSK0,43
+gengeneeval-0.3.0.dist-info/licenses/LICENSE,sha256=RDHgHDI4rSDq35R4CAC3npy86YUnmZ81ecO7aHfmmGA,1073
+gengeneeval-0.3.0.dist-info/RECORD,,