Question: Is parameter ordering requirement in SearchParam intentional?

[SHMAKI]

Hi biomass developers,

Thank you for this excellent library! I have a question about parameter handling in search_param.py.

What I observed

While working with biomass==0.14.0, I noticed that the order of parameters in self.idx_params seems to affect the behavior of get_obj_val across different models. Specifically:

1. I tested with multiple models (including built-in examples)
2. In search_param.py, when I rearranged the self.idx_params list order
3. Calling model.get_obj_val with the same parameter values produced different objective function values

This appears to be a general behavior affecting all models, not specific to any particular model.

My question

Is this ordering requirement intentional? I couldn't find documentation mentioning that parameters need to be listed in a specific order in search_param.py.

Potential impact

If this is indeed a requirement, it might be worth documenting clearly, as it can be quite puzzling to debug when users unknowingly define parameters in a different order across their models.

Possible solutions (if this is unintended)

One approach could be to auto-sort the parameter lists internally:

# in search_param.py
class SearchParam(object):
    def __init__(self):
        # User-defined order (for readability/organization)
        _idx_params = [
            C.param_b,  # Users can group parameters logically
            C.param_a,
            C.param_d,
            C.param_c,
            # ...
        ]
        # Library uses sorted order internally
        self.idx_params = sorted(_idx_params)

But I wanted to check with you first whether this behavior is by design or if it's something that could be improved.
Thanks for your time and consideration!

[himoto]

Thanks for raising this issue and for the clear description!

This behavior is expected. The order of self.idx_params is critical because BioMASS uses enumerate(self.idx_params) when constructing search bounds. For example:

for i, j in enumerate(self.idx_params):
    search_rgn[0, j] = search_param[i] * 0.1  # lower bound
    search_rgn[1, j] = search_param[i] * 10.0  # upper bound

Here, the index i (from enumeration) must match the position of the corresponding parameter value in search_param. If you change the order of self.idx_params, you effectively change which parameter each value in search_param is mapped to, which explains the different results you observed in get_obj_val.

To avoid unexpected behavior, it’s important to keep the parameter order consistent before and after parameter estimation.

That said, I agree this should be mentioned more explicitly in the documentation to help new users avoid confusion. We’ll look into clarifying this in the docs.

[himoto]

Updated docs: https://biomass-core.readthedocs.io/en/latest/tutorial/cfos.html#set-lower-upper-bounds-of-parameters-to-be-estimated

[SHMAKI]

@himoto
Thank you for your quick and very clear explanation.

The code example showing how enumerate(self.idx_params) is used makes the reason for this behavior perfectly clear. I now understand the importance of keeping the order consistent.

I really appreciate you updating the documentation so quickly! This will definitely help other users in the future.
Thanks again for your help and the great work on this library. I'll close this issue now.