[ENH] to_df to also return the gaussian parameters

[chapochn]

Hello,
It would be great if there would be an option so that the function to_df could also return the gaussian parameters. Wondering how difficult it would be to add that?

[TomDonoghue]

Hey @chapochn - do you mean extracting the gaussian parameters as opposed to extracting the peak parameters, as it does currently? That's probably quite doable - it would probably mean updating model_to_dict (which does the work of organizing extracted parameters) to be able to specify what to extract.

[chapochn]

Yes, that would be great!

[danieltomasz]

hi @chapochn my workaround before the update

import pandas as pd
import numpy as np
from specparam.core.funcs import infer_ap_func
from specparam.core.info import get_ap_indices


def specparam2pandas(fg):
    """
    Converts a SpectralGroupModel object into a pandas DataFrame, with peak parameters and
    corresponding aperiodic fit information.

    Args:
    -----
    fg : specpramGroup
        The SpectralGroupModel object containing the fitting results.

    Returns:
    --------
    peaks_df : pandas.DataFrame
        A DataFrame with the peak parameters and corresponding aperiodic fit information.
        The columns are:
        - 'CF': center frequency of each peak
        - 'PW': power of each peak
        - 'BW': bandwidth of each peak
        - 'error': fitting error of the aperiodic component
        - 'r_squared': R-squared value of the aperiodic fit
        - 'exponent': exponent of the aperiodic component
        - 'offset': offset of the aperiodic component
        - 'knee': knee parameter of the aperiodic component [if is initially present in the fg object]
    Notes:
    ------
    This function creates two DataFrames. The first DataFrame `specparam_aperiodic`
    contains the aperiodic fit information and is based on the `aperiodic_params`
    attribute of the SpectralGroupModel object. The columns are inferred using the
    `get_ap_indices()` and `infer_ap_func()` functions from the specparam package.
    The second DataFrame `peak_df` contains the peak parameters and is based on the
    `peak_params` attribute of the SpectralGroupModel object. The column names are renamed
    to match the headers of `fooof_aperiodic`, and the 'ID' column is cast to integer.
    The two DataFrames are then merged based on a shared 'ID' column.
    """

    specparam_aperiodic = (
        pd.DataFrame(
            fg.get_params("aperiodic_params"),
            columns=get_ap_indices(
                infer_ap_func(np.transpose(fg.get_params("aperiodic_params")))
            ),
        )
        .assign(error=fg.get_params("error"), r_squared=fg.get_params("r_squared"))
        .reset_index(names=["ID"])
    )
    return (
        pd.DataFrame(fg.get_params("peak_params"))  # prepare peaks dataframe
        .set_axis(["CF", "PW", "BW", "ID"], axis=1)  # rename cols
        .astype({"ID": int})
        .join(specparam_aperiodic.set_index("ID"), on="ID")
    )

[danieltomasz]

@TomDonoghue did you considered to add option to just convert fitResults object to dict or dataframe "loseless" - as it is, without specifying the number of peaks of bands of interest (when peak_org is None)?

[TomDonoghue]

@danieltomasz - is the output you're thinking of in such a case equivalent to set the peak_org to the maximum number of peaks? Then you would have the full set of results in the dataframe. I'm not sure I want that to be what happens if peak_org is None, but maybe we could add a special value like (e.g. peak_org='all') to do that as a convenience?

[danieltomasz]

@TomDonoghue yes, I meant maximum number of peak :) peak_org='all' will do the work, I am just recently using None as a default argument (then the function use the predefined parameters from the ) , but indeed Explicit is better than implicit. so `peak_org='all' is better;

I am able to use my custom functions to extract it, but as you plan to update model_to_dict this might be a good momement to think if it might benefit more users (current cases cover most of needs, but I prefer just to see/visualise all results with other python packages/R and then filter the the results to specific criteria)