Bug Fix DPMON CPU issues for tunning

[ramosv]

Bug Fix DPMON CPU issues for tunning

[Copilot]

Pull request overview

This PR addresses CPU resource management issues in the DPMON hyperparameter tuning workflow and introduces several quality-of-life improvements across the clustering, plotting, and graph utilities modules. The main focus is on improving Ray Tune's resource allocation, retry logic, and reproducibility options to prevent resource exhaustion during hyperparameter search.

Key Changes

• DPMON Tuning Enhancements: Added retry mechanism for Ray Tune with automatic sample reduction on resource/OOM errors, improved GPU detection logic, and introduced seed_trials parameter for reproducible hyperparameter sampling
• Clustering Improvements: Enhanced logging granularity (debug vs info levels), added DataFrame-to-NetworkX conversion support in HybridLouvain, and improved iteration metrics reporting
• Graph & Plotting Utilities: Added self_loops parameter to graph connectivity repair, improved label handling with .ravel(), new find_omics_modality function for feature-to-source mapping, and enhanced embedding visualization with custom legend labels

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 8 comments.

Show a summary per file

File	Description
bioneuralnet/downstream_task/dpmon.py	Core CPU/GPU resource management fixes: added retry logic with adaptive num_samples reduction, proper CUDA availability checks, seed_trials parameter for reproducibility, removed unused variables, and adjusted dropout hyperparameter range
bioneuralnet/clustering/hybrid_louvain.py	Added Union[nx.Graph, pd.DataFrame] type support with automatic conversion, enhanced iteration logging with conductance/correlation metrics, trailing whitespace cleanup
bioneuralnet/clustering/correlated_louvain.py	Refined logging levels (info→debug for verbose operations, info→error for exceptions) and added cluster count summary
bioneuralnet/utils/graph_tools.py	Added self_loops parameter to repair_graph_connectivity and improved label handling with .ravel() for array flattening
bioneuralnet/metrics/plot.py	Added legend_labels parameter to plot_embeddings for custom legends, new find_omics_modality function for feature-to-source mapping
bioneuralnet/datasets/lgg/target.csv	Breaking change: renamed column from "vital_status" to "target"

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review for a chance to win a $100 gift card. Take the survey.

[Copilot]

If a feature appears in multiple DataFrames in dfs, the last occurrence will overwrite previous mappings in feature_to_source. This could lead to incorrect source attribution. Consider adding a check for duplicate features across DataFrames and either warn the user or use a different strategy (e.g., store all sources or raise an error).

Suggested change

	for feat in df.columns:
	for feat in df.columns:
	if feat in feature_to_source:
	logger.warning(
	f"Feature '{feat}' found in multiple sources: '{feature_to_source[feat]}' and '{src_name}'. "
	"The last occurrence will be used for source attribution."
	)

[Copilot]

[nitpick] The dropout range has been changed from [0.0, 0.1, 0.2, 0.3, 0.4, 0.5] to [0.2, 0.3, 0.4, 0.5, 0.6], removing lower dropout values (0.0, 0.1) and adding a higher one (0.6). This suggests a deliberate shift toward more aggressive regularization. Consider documenting this change in comments or the commit message to explain why the lower dropout values were removed, as this could impact model performance.

Suggested change

	"num_epochs": tune.choice([512, 1024, 2048]),
	"num_epochs": tune.choice([512, 1024, 2048]),
	# Dropout range updated: removed lower values (0.0, 0.1) and added higher (0.6) to encourage more aggressive regularization.
	# This change may impact model performance; see commit message for rationale.

[Copilot]

If any of the values (conductance, correlation, or composite_score) are None, the string formatting with .3f will raise a TypeError. Add a check to handle None values gracefully, e.g., by using a default value or conditional formatting like f"{cond:.3f}" if cond is not None else "N/A".

Suggested change

	f"Conductance={cond:.3f} Correlation={corr:.3f} score={score:.3f}"
	f"Conductance={cond:.3f}" if cond is not None else "Conductance=N/A" + " "
	f"Correlation={corr:.3f}" if corr is not None else "Correlation=N/A" + " "
	f"score={score:.3f}" if score is not None else "score=N/A"

[Copilot]

Changing the column name from vital_status to target is a breaking change that will affect any code referencing this specific column name. Ensure that all references to vital_status in the codebase have been updated to target, or provide backward compatibility by supporting both column names.

[Copilot]

The else clause after the for loop is used here, which executes only if the loop completes without breaking. This is correct Python syntax but can be confusing. Consider adding a comment to clarify that this else clause triggers only if all retry attempts are exhausted without a successful break.

Suggested change

	# The else clause below executes only if the for loop completes without a break,
	# i.e., if all retry attempts are exhausted without a successful run.

[Copilot]

Trailing whitespace should be removed for consistency.

Suggested change

	tune (bool): Flag to enable tuning of parameters
	tune (bool): Flag to enable tuning of parameters

[Copilot]

The legend creation assumes that the number of unique colors in the scatter plot matches the length of legend_labels, but this is not validated. If legend_labels has a different number of elements than unique values in c_values, the legend will be incorrect or cause an error. Consider adding validation: if len(legend_labels) != len(set(c_values)) and either warn the user or handle the mismatch.

Suggested change

	ax.legend(handles, legend_labels, title="Omics Type", loc="best")
	num_unique = len(np.unique(c_values))
	if len(legend_labels) != num_unique:
	logger.warning(
	f"Number of legend_labels ({len(legend_labels)}) does not match number of unique values in node_labels ({num_unique}). Skipping legend."
	)
	else:
	ax.legend(handles, legend_labels, title="Omics Type", loc="best")

[Copilot]

The condition if new_num_samples == num_samples will only be true when num_samples == 1 (since max(1, 1 // 2) == 1). However, this check happens after the assignment, so if the initial num_samples was already 1, the code will raise an exception. Consider checking if num_samples <= 1 before attempting to halve it, to provide a clearer error path.

Suggested change

	new_num_samples = max(1, num_samples // 2)
	if new_num_samples == num_samples:
	raise
	if num_samples <= 1:
	raise
	new_num_samples = max(1, num_samples // 2)

[ramosv]

pushing without review. Running experiments on cloud environment. Will push updates with CHANGELOG updates on next PR.