Is it possible that parameter in the drift rate also be used for other parameter estimation like boundary or noise

[zaizibai]

Dear Pyddm developer,

Thanks for making this wonderful tool. I am trying to use the decision field theory from Dai & Busemeyer, 2014 in pyddm. In this model, drift rate is defined as: d = w*(u(v_l)-u(v_s)) - (1-w)(p_(t_l)-p(t_s)). v_l, v_s, t_l and t_s are the experiment variable. w and the parameter in the utility function u(x) and p(x) are free parameters. In this model, the noise of evidence accumulation is noise = \sqrt{w(u(v_l)-u(v_s)))^2 + (1-w) *(p_(t_l)-p(t_s))^2 - d^2 }. Parameter in the drift rate are also used in the noise parameter. Is it possible to build this DFT model in pyddm? Thanks a lot.

Best,
Mingqian

[mwshinn]

From what you've written here, yes, it sounds possible!

[zaizibai]

If i delete the noisenp.sqrt(w_r * (amountp_r -20p_r)**2+w_t(Delayp_t)**2), the the model-fitting is fine.

[mwshinn]

If you would like help, please make a minimal reproducible example.

[zaizibai]

If you would like help, please make a minimal reproducible example.

Hi,
Thanks for the information. Here i made a minimal example:

model = pyddm.gddm(
    name = 'ddm_test',
    drift=lambda v_0, v_1, cond: v_0 + v_1 * cond ,
     bound=lambda b_0,b_1,v_0, v_1, cond: b_0 +b_1 * np.abs(v_1 * cond) ,
    starting_position="z",
    noise = 1,
    nondecision="ndt",
    mixture_coef=.05,
   parameters={"v_0": (-5,5),"v_1":(-5,5),
              "b_0":(-3,3), "b_1":(-3,3),
               "ndt":(.01,3),"z":(-.6,.6)},
    dx=.01, dt=.001,
   conditions=["cond"])

I would like to allow one part of the drift rate (v_1 * cond) also modulates the boundary parameter. Looking forward to your reply. Thanks a lot.

[mwshinn]

In the model you posted, the v_1 * cond will indeed modulate both drift rate and bound. However this model will probably cause errors because the bound can be negative (since b_0 and b_1 can be negative). If you constrain b_0 and b_1 to be greater than 0, or else set bound=lambda b_0,b_1,v_0, v_1, cond: np.maximum(.1, b_0 +b_1 * np.abs(v_1 * cond)) to ensure the bound is strictly positive, then it should work! (Though also, you don't need bound to be a function of v_0 since you don't use that anywhere in the bound equation.)