[Legacy v1.9.16] evaluate_components crashes with TypeError when using seeded initialization (Ain) and only_init=True

[LinxiuZeng]

I am using a legacy local version of CaImAn (v1.9.16) for longitudinal consistency in an endoscopic (CNMF-E) processing pipeline. I am implementing an iterative refinement strategy where spatial and temporal estimates from a first pass are used to seed a second pass.

When providing Ain, Cin, b_in, and f_in to the CNMF object and setting only_init=True (following the CNMF-E demo logic), the fit() function completes but leaves the estimates.YrA attribute as None.

When calling evaluate_components() immediately after, the process crashes because the evaluation logic attempts to sum the traces and residuals (C + YrA).

Minimal Reproducible Example (v1.9.16)

# Seeded Initialization for CNMF-E
# Ain, Cin, b_in, and f_in are all from the last cnmfe pass
cnm = cnmf.CNMF(n_processes=n_processes, 
                dview=dview, 
                Ain=iterativeA, 
                Cin=iterativeC, 
                b_in=iterativeb, 
                f_in=iterativef, 
                params=opts)

# opts has 'only_init': True and 'method_init': 'corr_pnr' (or 'seeded')
cnm.fit(images) 

# Result: cnm.estimates.YrA is None
# This call fails:
cnm.estimates.evaluate_components(images, cnm.params, dview=dview)

Traceback

File "caiman/components_evaluation.py", line 608, in estimate_components_quality_auto
    traces = C + YrA
TypeError: unsupported operand type(s) for +: 'float' and 'NoneType'

Questions

1. In v1.9.16, is it expected that only_init=True skips the residual (YrA) calculation even when Ain is provided?
2. To maintain the CNMF-E model while achieving iterative refinement, is the recommended workaround to manually call compute_residuals() before evaluation, or should only_init be set to False for the seeded run?
3. In the official demo_pipeline_cnmfE.ipynb, only_init is set to True (Cell 11); I am concerned that changing this to False to resolve the YrA/evaluation crash might bypass the specialized CNMF-E local background model in favor of standard CNMF refinement.
4. Is there a specific parameter configuration in v1.9.16 that ensures the background components are properly updated during a seeded refinement pass without reverting to the CNMF model?

Environment
CaImAn version: 1.9.16
Installation: Local source files (.../Downloads..../CaImAn-main/...)
Data Type: 1-photon/CNMF-E