omnigenome 0.3.24a0__tar.gz → 0.4.0a0__tar.gz

{omnigenome-0.3.24a0 → omnigenome-0.4.0a0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: omnigenome
-Version: 0.3.24a0
+Version: 0.4.0a0
 Summary: OmniGenome: A comprehensive toolkit for genome analysis.
 Home-page: https://github.com/yangheng95/OmniGenBench
 Author: Yang, Heng
@@ -182,7 +182,7 @@ ogb autobench \
     --trainer accelerate
 
 # Legacy command (still supported for backward compatibility)
-# autobench --model_name_or_path "yangheng/OmniGenome-186M" --benchmark "RGB"
+# autobench --config_or_model "yangheng/OmniGenome-186M" --benchmark "RGB"
 ```
 **Output**: Results include mean ± standard deviation for each metric (e.g., MCC: 0.742 ± 0.015, F1: 0.863 ± 0.009)
 
@@ -202,7 +202,7 @@ seeds = [0, 1, 2, 3, 4]  # Multi-seed for statistical rigor
 # Run automated evaluation
 bench = AutoBench(
     benchmark=benchmark,
-    model_name_or_path=gfm,
+    config_or_model=gfm,
     overwrite=False  # Skip completed tasks
 )
 bench.run(autocast=False, batch_size=bench_size, seeds=seeds)
@@ -327,7 +327,7 @@ RNA secondary structure prediction is a fundamental problem in computational bio
 where the goal is to predict the secondary structure of an RNA sequence.
 In this demo, we show how to use OmniGenBench to predict the secondary structure of RNA sequences using a pre-trained model.
 The tutorials of RNA Secondary Structure Prediction can be found in
-[Secondary_Structure_Prediction_Tutorial.ipynb](examples/rna_secondary_structure_prediction/Secondary_Structure_Prediction_Tutorial.ipynb).
+[Secondary_Structure_Prediction_Tutorial.ipynb](examples/rna_secondary_structure_prediction/00_quickstart_rna_ssp.ipynb)(examples/rna_secondary_structure_prediction/00.ipynb).
 
 You can find a visual example of RNA Secondary Structure Prediction [here](asset/RNASSP-Demo.gif).
 

{omnigenome-0.3.24a0 → omnigenome-0.4.0a0}/omnigenome/__init__.py

@@ -29,7 +29,7 @@ import warnings
 
 warnings.warn(
     "The 'omnigenome' package is deprecated, please use omnigenbench package instead. "
-    "e.g., from omnigenome import *  ->  from omnigenbench import *\n"
+    "e.g., from omnigenbench import *  ->  from omnigenbench import *\n"
     "All imports from omnigenome will be redirected to omnigenbench. ",
     DeprecationWarning,
 )
@@ -59,6 +59,7 @@ try:
         OmniDatasetForSequenceRegression,
         OmniDatasetForTokenClassification,
         OmniDatasetForTokenRegression,
+        OmniDatasetForMultiLabelClassification,
     )
 
     # Import metric classes
@@ -99,6 +100,15 @@ try:
         OmniModelForAugmentation,
     )
 
+    from omnigenbench.src.model.baselines import (
+        OmniCNNBaseline,
+        OmniRNNBaseline,
+        OmniBPNetBaseline,
+        OmniBasenjiBaseline,
+        OmniDeepSTARRBaseline,
+        OmniGenericBaseline,
+    )
+
     # Import LoRA model
     from omnigenbench.src.lora.lora_model import OmniLoraModel
 
@@ -125,17 +135,27 @@ try:
 
     # Import hub classes
     from omnigenbench.src.utility.model_hub.model_hub import ModelHub
-    from omnigenbench.src.utility.dataset_hub import load_benchmark_datasets
-    from omnigenbench.src.utility.pipeline_hub import Pipeline
+    from omnigenbench.src.utility.dataset_hub.dataset_hub import load_benchmark_datasets
+    from omnigenbench.src.utility.pipeline_hub.pipeline import Pipeline
     from omnigenbench.src.utility.pipeline_hub.pipeline_hub import PipelineHub
 
     # Import module utilities
     from omnigenbench.src.model.module_utils import OmniPooling
-    from omnigenbench.src.utility import VoteEnsemblePredictor
+    from omnigenbench.src.utility.ensemble import VoteEnsemblePredictor
 
     # For backward compatibility version 0.2.7alpha and earlier
     from omnigenbench.auto.config.auto_config import AutoBenchConfig
 
+    # Import explainer classes
+    from omnigenbench.src.explainability.epistasis.explainer import EpistasisExplainer
+    from omnigenbench.src.explainability.sequence_logo.explainer import (
+        SequenceLogoExplainer,
+    )
+    from omnigenbench.src.explainability.visualization_2d.explainer import (
+        Visualization2DExplainer,
+    )
+    from omnigenbench.src.explainability.attention.explainer import AttentionExplainer
+
     # Create backward compatibility aliases
     OmniGenomeTokenizer = OmniTokenizer
     OmniGenomeKmersTokenizer = OmniKmersTokenizer
@@ -167,6 +187,7 @@ try:
 
     # Define __all__ for explicit exports
     __all__ = [
+        "__version__",
         "load_benchmark_datasets",
         "OmniDataset",
         "OmniModel",
@@ -203,6 +224,44 @@ try:
         "print_args",
         "env_meta_info",
         "RNA2StructureCache",
+        "OmniDatasetForSequenceClassification",
+        "OmniDatasetForSequenceRegression",
+        "OmniDatasetForTokenClassification",
+        "OmniDatasetForTokenRegression",
+        "OmniDatasetForMultiLabelClassification",
+        "OmniTokenizer",
+        "OmniKmersTokenizer",
+        "OmniSingleNucleotideTokenizer",
+        "OmniBPETokenizer",
+        "OmniDataset",
+        "OmniMetric",
+        "OmniModel",
+        "OmniLoraModel",
+        "OmniModelForSequenceClassification",
+        "OmniModelForMultiLabelSequenceClassification",
+        "OmniModelForTokenClassification",
+        "OmniModelForSequenceRegression",
+        "OmniModelForTokenRegression",
+        "OmniModelForStructuralImputation",
+        "OmniModelForMatrixRegression",
+        "OmniModelForMatrixClassification",
+        "OmniModelForMLM",
+        "OmniModelForSeq2Seq",
+        "OmniModelForRNADesign",
+        "OmniModelForEmbedding",
+        "OmniModelForAugmentation",
+        "OmniPooling",
+        "download_benchmark",
+        "download_model",
+        "download_pipeline",
+        "query_models_info",
+        "hub_utils",
+        "OmniCNNBaseline",
+        "OmniRNNBaseline",
+        "OmniBPNetBaseline",
+        "OmniBasenjiBaseline",
+        "OmniDeepSTARRBaseline",
+        "OmniGenericBaseline",
         # OmniGenome* aliases for backward compatibility
         "OmniGenomeTokenizer",
         "OmniGenomeKmersTokenizer",
@@ -234,19 +293,12 @@ try:
         "bench_command",
         "run_train",
         "train_command",
+        "EpistasisExplainer",
+        "SequenceLogoExplainer",
+        "Visualization2DExplainer",
+        "AttentionExplainer",
     ]
 
 except ImportError as e:
-    import warnings
-
-    warnings.warn(
-        f"Failed to import omnigenbench modules: {e}. "
-        "Please ensure omnigenbench is properly installed.\n"
-        "You can install it with: pip install omnigenbench\n"
-        "and replace all 'omnigenome' with 'omnigenbench' in your code.\n"
-        "e.g., from omnigenome import *  ->  from omnigenbench import *",
-        ImportWarning,
-    )
-
     # Minimal fallback to prevent complete failure
     __all__ = []

{omnigenome-0.3.24a0 → omnigenome-0.4.0a0}/omnigenome.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: omnigenome
-Version: 0.3.24a0
+Version: 0.4.0a0
 Summary: OmniGenome: A comprehensive toolkit for genome analysis.
 Home-page: https://github.com/yangheng95/OmniGenBench
 Author: Yang, Heng
@@ -182,7 +182,7 @@ ogb autobench \
     --trainer accelerate
 
 # Legacy command (still supported for backward compatibility)
-# autobench --model_name_or_path "yangheng/OmniGenome-186M" --benchmark "RGB"
+# autobench --config_or_model "yangheng/OmniGenome-186M" --benchmark "RGB"
 ```
 **Output**: Results include mean ± standard deviation for each metric (e.g., MCC: 0.742 ± 0.015, F1: 0.863 ± 0.009)
 
@@ -202,7 +202,7 @@ seeds = [0, 1, 2, 3, 4]  # Multi-seed for statistical rigor
 # Run automated evaluation
 bench = AutoBench(
     benchmark=benchmark,
-    model_name_or_path=gfm,
+    config_or_model=gfm,
     overwrite=False  # Skip completed tasks
 )
 bench.run(autocast=False, batch_size=bench_size, seeds=seeds)
@@ -327,7 +327,7 @@ RNA secondary structure prediction is a fundamental problem in computational bio
 where the goal is to predict the secondary structure of an RNA sequence.
 In this demo, we show how to use OmniGenBench to predict the secondary structure of RNA sequences using a pre-trained model.
 The tutorials of RNA Secondary Structure Prediction can be found in
-[Secondary_Structure_Prediction_Tutorial.ipynb](examples/rna_secondary_structure_prediction/Secondary_Structure_Prediction_Tutorial.ipynb).
+[Secondary_Structure_Prediction_Tutorial.ipynb](examples/rna_secondary_structure_prediction/00_quickstart_rna_ssp.ipynb)(examples/rna_secondary_structure_prediction/00.ipynb).
 
 You can find a visual example of RNA Secondary Structure Prediction [here](asset/RNASSP-Demo.gif).
 

{omnigenome-0.3.24a0 → omnigenome-0.4.0a0}/omnigenome.egg-info/SOURCES.txt

@@ -9,8 +9,17 @@ omnigenome.egg-info/entry_points.txt
 omnigenome.egg-info/requires.txt
 omnigenome.egg-info/top_level.txt
 tests/test_attention_extraction.py
+tests/test_autobench_autotrain.py
+tests/test_autobench_hub_integration.py
 tests/test_autoinfer_cli.py
+tests/test_autotrain_hub_integration.py
+tests/test_benchmark_download.py
+tests/test_cli_commands.py
+tests/test_cli_parameter_mapping.py
+tests/test_example_notebooks.py
 tests/test_genomic_embeddings.py
+tests/test_hf_download.py
 tests/test_rna_design.py
 tests/test_structure_prediction.py
-tests/test_token_classification.py
+tests/test_token_classification.py
+tests/test_training_workflows.py

{omnigenome-0.3.24a0 → omnigenome-0.4.0a0}/setup.py

@@ -1,5 +1,5 @@
 # -*- coding: utf-8 -*-
-# file: setup.py
+# file: setup_omnigenbench.py
 # time: 14:54 06/04/2024
 # author: YANG, HENG <hy345@exeter.ac.uk> (杨恒)
 # github: https://github.com/yangheng95
@@ -35,12 +35,10 @@ extras = {
     ]
 }
 
-# This is the main setup.py - it will build omnigenbench by default
-# Use setup_omnigenome.py and setup_omnigenbench.py for separate builds
 setup(
     name="omnigenbench",
     version=read_version_from_init(),
-    description="OmniGenoBench: A comprehensive toolkit for genome analysis.",
+    description="OmniGenBench: A comprehensive toolkit for genome analysis benchmarking.",
     long_description=long_description,
     long_description_content_type="text/markdown",
     url="https://github.com/yangheng95/OmniGenBench",
@@ -51,14 +49,12 @@ setup(
     include_package_data=True,
     exclude_package_data={"": [".gitignore"]},
     license="Apache-2.0",
-    packages=find_packages(include=["omnigenbench", "omnigenbench.*", "omnigenome", "omnigenome.*"]),
+    packages=find_packages(include=["omnigenbench", "omnigenbench.*"]),
     entry_points={
         "console_scripts": [
-            "ogb=omnigenbench.cli.ogb_cli:main",
-            # Legacy commands for backward compatibility
             "autobench=omnigenbench.auto.auto_bench.auto_bench_cli:run_bench",
             "autotrain=omnigenbench.auto.auto_train.auto_train_cli:run_train",
-            "autoinfer=omnigenbench.cli.autoinfer_cli:main",
+            "ogb=omnigenbench.cli.ogb_cli:main",
         ],
     },
     install_requires=[
@@ -76,7 +72,6 @@ setup(
         "packaging",
         "peft",
         "dill",
-        "accelerate",
         "plotly",
         "logomaker",
         "matplotlib",

{omnigenome-0.3.24a0 → omnigenome-0.4.0a0}/setup_omnigenome.py

@@ -11,7 +11,7 @@ from pathlib import Path
 from setuptools import setup, find_packages
 
 # Define version directly to avoid circular import
-from omnigenome import __version__
+from omnigenbench import  __version__
 
 cwd = Path(__file__).parent
 long_description = (cwd / "README.MD").read_text(encoding="utf8")

{omnigenome-0.3.24a0 → omnigenome-0.4.0a0}/tests/test_attention_extraction.py

@@ -46,7 +46,8 @@ class TestAttentionExtractionEmbeddingModel:
     @pytest.fixture(scope="class")
     def embedding_model(self, model_name):
         """Load embedding model for attention extraction"""
-        model = OmniModelForEmbedding(model=model_name, trust_remote_code=True)
+        # OmniModelForEmbedding takes config_or_model as first positional argument
+        model = OmniModelForEmbedding(model_name, trust_remote_code=True)
         return model
 
     def test_single_sequence_attention_extraction(self, embedding_model, test_sequences):
@@ -164,7 +165,7 @@ class TestAttentionExtractionBatch:
     @pytest.fixture(scope="class")
     def embedding_model(self, model_name):
         """Load embedding model for batch extraction"""
-        model = OmniModelForEmbedding(model=model_name, trust_remote_code=True)
+        model = OmniModelForEmbedding(model_name, trust_remote_code=True)
         return model
 
     def test_batch_attention_extraction(self, embedding_model, test_sequences):
@@ -229,43 +230,83 @@ class TestAttentionExtractionTaskModels:
     def test_classification_model_attention(self, model_name, test_sequences):
         """Test attention extraction from classification model"""
         # Use classification model (also supports attention extraction)
+        # Need to load tokenizer first for classification models
+        from omnigenbench import OmniTokenizer
+        tokenizer = OmniTokenizer.from_pretrained(model_name)
+
+        # Classification model requires config_or_model and tokenizer as positional args
         model = OmniModelForSequenceClassification(
-            model=model_name,
+            model_name,
+            tokenizer,
             num_labels=2,
-            trust_remote_code=True
+            trust_remote_code=True,
         )
-        
+
+        # Some installed versions may not expose EmbeddingMixin on task models
+        if not hasattr(model, "extract_attention_scores"):
+            pytest.xfail(
+                "Installed omnigenbench version does not expose attention extraction on task models; "
+                "this is available in newer local source."
+            )
+
+        # Ensure device attribute exists for EmbeddingMixin in older builds
+        if not hasattr(model, "device"):
+            model.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
         sequence = test_sequences[0]
         attention_result = model.extract_attention_scores(
             sequence=sequence,
             max_length=128,
-            return_on_cpu=True
+            return_on_cpu=True,
         )
-        
-        assert "attentions" in attention_result, \
+
+        assert "attentions" in attention_result, (
             "Classification model should support attention extraction"
-        assert isinstance(attention_result['attentions'], torch.Tensor), \
+        )
+        assert isinstance(attention_result["attentions"], torch.Tensor), (
             "Should return attention tensor"
+        )
 
     def test_regression_model_attention(self, model_name, test_sequences):
         """Test attention extraction from regression model"""
         # Use regression model (also supports attention extraction)
+        # Need to load tokenizer first for regression models
+        from omnigenbench import OmniTokenizer
+        tokenizer = OmniTokenizer.from_pretrained(model_name)
+
+        # Regression model requires config_or_model and tokenizer as positional args
+        # Also requires num_labels or label2id; for regression use 1 output
         model = OmniModelForSequenceRegression(
-            model=model_name,
-            trust_remote_code=True
+            model_name,
+            tokenizer,
+            num_labels=1,
+            trust_remote_code=True,
         )
-        
+
+        # Some installed versions may not expose EmbeddingMixin on task models
+        if not hasattr(model, "extract_attention_scores"):
+            pytest.xfail(
+                "Installed omnigenbench version does not expose attention extraction on task models; "
+                "this is available in newer local source."
+            )
+
+        # Ensure device attribute exists for EmbeddingMixin in older builds
+        if not hasattr(model, "device"):
+            model.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
         sequence = test_sequences[0]
         attention_result = model.extract_attention_scores(
             sequence=sequence,
             max_length=128,
-            return_on_cpu=True
+            return_on_cpu=True,
         )
-        
-        assert "attentions" in attention_result, \
+
+        assert "attentions" in attention_result, (
             "Regression model should support attention extraction"
-        assert isinstance(attention_result['attentions'], torch.Tensor), \
+        )
+        assert isinstance(attention_result["attentions"], torch.Tensor), (
             "Should return attention tensor"
+        )
 
 
 class TestAttentionExtractionEdgeCases:
@@ -274,7 +315,7 @@ class TestAttentionExtractionEdgeCases:
     @pytest.fixture(scope="class")
     def embedding_model(self, model_name):
         """Load embedding model"""
-        model = OmniModelForEmbedding(model=model_name, trust_remote_code=True)
+        model = OmniModelForEmbedding(model_name, trust_remote_code=True)
         return model
 
     def test_very_short_sequence(self, embedding_model):
@@ -343,7 +384,7 @@ class TestAttentionExtractionPerformance:
     @pytest.fixture(scope="class")
     def embedding_model(self, model_name):
         """Load embedding model"""
-        model = OmniModelForEmbedding(model=model_name, trust_remote_code=True)
+        model = OmniModelForEmbedding(model_name, trust_remote_code=True)
         return model
 
     def test_large_batch_processing(self, embedding_model):