export-to-phy as standalone feature, allows compute of PCA features from raw files

[shrutisss-21]

Added run_export_to_phy, so that you can run reanalyze_bursts and export-to-phy if you didnt run it originally. it will also compute pca features using original h5 file if you deleted your binary file

[Copilot]

Pull request overview

Adds a standalone “export-to-phy” path to the MEA pipeline so users can re-run burst analysis and generate phy_output even if Phy export wasn’t done originally, including an attempt to recompute PCA features by reloading the original .h5 when the binary cache is missing.

Changes:

• Change generate_reports() defaults and update its export_to_phy call to explicitly disable PC feature computation.
• Add MEAPipeline.run_export_to_phy() to export a curated unit set to Phy, optionally attaching a recording from binary/ or the original .h5 for PCA computation.
• In run_mea_pipeline(), when reanalyze_bursts is used, optionally generate phy_output if --export-to-phy was requested and the folder doesn’t already exist.

---

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

[Copilot]

This change forces compute_pc_features=False for exports triggered from generate_reports, which disables PCA features even when the binary cache/recording is available (the previous behavior relied on SpikeInterface defaults). If the goal is only to avoid failures when the binary has been deleted, consider determining this flag based on whether a recording is attached/available, or reusing the new run_export_to_phy logic instead of hard-disabling PC features here.

Suggested change

	si.export_to_phy(
	self.analyzer.select_units(clean_units),
	output_folder=self.output_dir / "phy_output",
	remove_if_exists=True,
	copy_binary=False,
	compute_pc_features = False # cannot generate pc features without original binary file
	)
	# Use the standalone export helper so PC feature computation
	# is determined based on recording/binary availability.
	self.run_export_to_phy(clean_units)

[Copilot]

run_export_to_phy sets compute_pc_features=True when self.recording is present, but if self.analyzer was loaded from disk earlier via load_sorting_analyzer(analyzer_folder) it may not have a recording attached (load_recording defaults to False). In that case export_to_phy(..., compute_pc_features=True) can fail or silently skip PC computation. Consider ensuring the analyzer has a recording when compute_pc_features is True (e.g., reload with load_recording=True when binary exists, or call set_temporary_recording(self.recording)).

Suggested change

	compute_pc_features = True
	try:
	# If the analyzer was loaded from disk without a recording, attach the in-memory one.
	if hasattr(self.analyzer, "get_recording") and self.analyzer.get_recording() is None:
	self.analyzer.set_temporary_recording(self.recording)
	compute_pc_features = True
	except Exception as e:
	self.logger.warning(
	"Could not attach in-memory recording for PCA (%s) — will try other sources.",
	e,
	)

[Copilot]

run_export_to_phy always attempts to re-apply curation via _apply_curation_logic and has no way to mirror the existing no_curation option from generate_reports/MEARunOptions. This can unexpectedly export fewer units than the prior run (or than requested). Consider adding a no_curation (or export_all_units) parameter and skipping the curation block when requested, and pass it through from run_mea_pipeline.

[mandarmp]

@shrutisss-21 there are a few copilot suggestions which I need to look into before merging.