7-parameter ddm in HSSM possible?

[Franzi2114]

Hello together,

is it possible to estimate the full 7-parameter diffusion model in HSSM? If yes, how?

Best,
Franziska

[digicosmos86]

Hi Franziska,

It is possible through blackbox likelihoods if you are not using any hierarchical models at the moment. Happy to provide an example if that's if you want. We will be adding hierarchical model support for full_ddm models via blackbox likelihood soon.

[Franzi2114]

Hi Paul,
interesting! It would be nice to see an example how this works =D

[digicosmos86]

Here's an example in our tutorial. In this example, we basically use the full HDDM likelihood in Cython as a blackbox likelihood function. You can modify this example like this to make it work for you:

# You need to do pip install git+https://github.com/brown-ccv/hddm-wfpt.git first to install 
# the hddm likelihood functions.

import hddm_wfpt
import bambi as bmb

# Define a function with fun(data, *) signature
def my_blackbox_loglik(data, v, sv, a, z, sz, t, st, err=1e-8):
    data = data[:, 0] * data[:, 1]

    # Our function expects inputs as float64, but they are not guaranteed to
    # come in as such --> we type convert
    return hddm_wfpt.wfpt.pdf_array(
        np.float64(data),
        np.float64(v),
        np.float64(sv),
        np.float64(2 * a),
        np.float64(z),
        np.float64(sz),
        np.float64(t),
        np.float64(st),
        err,
        1,
    )


# Create the model with pdf_ddm_blackbox
model = hssm.HSSM(
    data=data,
    model="full_ddm",
    loglik=my_blackbox_loglik,
    loglik_kind="blackbox",
    model_config={
        "list_params": ["v", "sv", "a", "z", "sz", "t", "st"],
        "bounds": {
            "v": (-10.0, 10.0),
            "a": (0.0, 4.0),
            "z": (0.0, 1.0),
            "t": (0.0, 2.0), # Feel free to add additional bounds
        }
    },
    t=bmb.Prior("Uniform", lower=0.0, upper=2.0, initval=0.1),
)

sample = model.sample()

[Franzi2114]

Great, thanks! So it's also possible to set priors for all model parameters now?

[digicosmos86]

Yes, please refer to this guide to specify all priors

[frankmj]

we may also add a LAN or other approximate likelihood for the full DDM model as that is much faster than doing numerical integration (see Fengler eLife paper) - but requires some more work to properly span the parameter space and validate before including in HSSM. Michael

…

On Fri, Aug 4, 2023 at 11:32 AM Paul Xu ***@***.***> wrote: Hi Franziska, It is possible through blackbox likelihoods if you are not using any hierarchical models at the moment. Happy to provide an example if that's if you want. We will be adding hierarchical model support for full_ddm models via blackbox likelihood soon. — Reply to this email directly, view it on GitHub <#255 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAG7TFB46J2ELIYLBL3A733XTUIZVANCNFSM6AAAAAA3D6LJ6I> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>