search

lnccbrown / HSSM

Development of HSSM package
Other
70 stars 10 forks source link

0.2.2 might have broken the race _4 model #463

Closed hyang336 closed 2 weeks ago

hyang336 commented 2 weeks ago

Describe the bug When running the same model that worked on 0.2.1, the following error occurred:

Traceback (most recent call last): File "/home/hyang336/PPCPRC/analyses/test_phase/HSSM/Simulations.py", line 199, in model_race4nba_v_true = hssm.HSSM( ^^^^^^^^^^ File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/hssm/hssm.py", line 383, in init self._post_check_data_sanity() File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/hssm/hssm.py", line 1685, in _post_check_data_sanity raise ValueError( ValueError: Invalid responses found in your dataset: [0, 2, 3]

HSSM version 0.2.2

To Reproduce a0=0 b0=1 intercept0=0.85

a1=1.5
b1=3
intercept1=2.1

a2=3
b2=1.5
intercept2=2.1

a3=1
b3=0
intercept3=0.85

n_subjects=30 #number of subjects
n_trials=200 #number of trials per subject
param_sv=0.2 #standard deviation of the subject-level parameters

subject_params={
    "v0": np.array([]),
    "v1": np.array([]),
    "v2": np.array([]),
    "v3": np.array([]),
    "simneural": np.array([]),
    "subID": np.array([])
}

sim_data=[]

for i in range(n_subjects):
    # set the seed for each subject deterministically so all models are based on the same data
    np.random.seed(i)
    # generate neural data
    simneural=np.random.uniform(0, 1, size=n_trials)
    # generate v0, v1, v2, v3
    v0=np.exp(np.random.normal(intercept0, param_sv) + a0*np.log(simneural) + b0*np.log(1-simneural))
    v1=np.exp(np.random.normal(intercept1, param_sv) + a1*np.log(simneural) + b1*np.log(1-simneural))
    v2=np.exp(np.random.normal(intercept2, param_sv) + a2*np.log(simneural) + b2*np.log(1-simneural))
    v3=np.exp(np.random.normal(intercept3, param_sv) + a3*np.log(simneural) + b3*np.log(1-simneural))

v0_inb= np.where(np.logical_and(v0>= 0,v0<= 2.5))
v1_inb= np.where(np.logical_and(v1>= 0,v1<= 2.5))
v01_inb=np.intersect1d(v0_inb[0],v1_inb[0])

v2_inb= np.where(np.logical_and(v2>= 0,v2<= 2.5))
v3_inb= np.where(np.logical_and(v3>= 0,v3<= 2.5))
v23_inb=np.intersect1d(v2_inb[0],v3_inb[0])

v0123_inb=np.intersect1d(v01_inb,v23_inb)#indices of elements that are in bound for all 4 arrays

simneural=simneural[v0123_inb]
v0=v0[v0123_inb]
v1=v1[v0123_inb]
v2=v2[v0123_inb]
v3=v3[v0123_inb]

subject_params["v0"]=np.append(subject_params["v0"],v0)
subject_params["v1"]=np.append(subject_params["v1"],v1)
subject_params["v2"]=np.append(subject_params["v2"],v2)
subject_params["v3"]=np.append(subject_params["v3"],v3)
subject_params["simneural"]=np.append(subject_params["simneural"],simneural)
subject_params["subID"]=np.append(subject_params["subID"],np.repeat(i,len(simneural)))

true_values = np.column_stack(
[v0,v1,v2,v3, np.repeat([[2.0, 0.0, 1e-3,0.0]], axis=0, repeats=len(simneural))]
)

race4nba_v = simulator.simulator(true_values, model="race_no_bias_angle_4", n_samples=1)

rand_x = np.random.uniform(0, 1, size=len(simneural))
sim_data.append(
    pd.DataFrame(
        {
            "rt": race4nba_v["rts"].flatten(),
            "response": race4nba_v["choices"].flatten(),
            "x": np.log(simneural),
            "y": np.log(1-simneural),
            "rand_x": np.log(rand_x),
            "rand_y": np.log(1-rand_x),
            "subID": i
        }
    )
)

sim_data_concat=pd.concat(sim_data)

model_race4nba_v_true = hssm.HSSM(
data=sim_data_concat,
model='race_no_bias_angle_4',
a=2.0,
z=0.0,
include=[
    {
        "name": "v0",
        "prior":{"name": "Uniform", "lower": 0, "upper": 2.5},
        "formula": "v0 ~ 1 + x + y + (1|subID)",
        "link": "log",
    },
    {
        "name": "v1",
        "prior":{"name": "Uniform", "lower": 0, "upper": 2.5},
        "formula": "v1 ~ 1 + x + y + (1|subID)",
        "link": "log",
    },
    {
        "name": "v2",
        "prior":{"name": "Uniform", "lower": 0, "upper": 2.5},
        "formula": "v2 ~ 1 + x + y + (1|subID)",
        "link": "log",
    },
    {
        "name": "v3",
        "prior":{"name": "Uniform", "lower": 0, "upper": 2.5},
        "formula": "v3 ~ 1 + x + y + (1|subID)",
        "link": "log",
    }
],
)

Screenshots NA

Additional context Add any other context about the problem here, such as the print out of the model.

digicosmos86 commented 2 weeks ago

Hello,

In HSSM 0.2.2, we added the ability for the users to specify the number of choices in the HSSM class. You can use the “choices” argument to specify the number of choices that you have. We did it so that we can more robustly check the issues in the dataset the users provide.

Let me know if this works for you. Thanks!

http://ccv.brown.edu/ Paul Xu, PhD https://github.com/digicosmos86

Lead Data Scientist Center for Computation and Visualization Brown University | OIT Pronouns: he/him/his

On Sat, Jun 15, 2024 at 7:43 PM hyang336 @.***> wrote:

Describe the bug When running the same model that worked on 0.2.1, the following error occurred:

Traceback (most recent call last): File "/home/hyang336/PPCPRC/analyses/test_phase/HSSM/Simulations.py", line 199, in model_race4nba_v_true = hssm.HSSM( ^^^^^^^^^^ File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/hssm/hssm.py", line 383, in init self._post_check_data_sanity() File "/home/hyang336/HSSM022_tempENV/lib/python3.11/site-packages/hssm/hssm.py", line 1685, in _post_check_data_sanity raise ValueError( ValueError: Invalid responses found in your dataset: [0, 2, 3]

HSSM version 0.2.2

To Reproduce a0=0 b0=1

intercept0=np.log(1/beta(a0,b0))

intercept0=0.85

a1=1.5 b1=3

intercept1=np.log(1/beta(a1,b1))

intercept1=2.1

a2=3 b2=1.5

intercept2=np.log(1/beta(a2,b2))

intercept2=2.1

a3=1 b3=0

intercept3=np.log(1/beta(a3,b3))

intercept3=0.85

n_subjects=30 #number of subjects n_trials=200 #number of trials per subject param_sv=0.2 #standard deviation of the subject-level parameters Save trial-level parameters for each subject

subject_params={ "v0": np.array([]), "v1": np.array([]), "v2": np.array([]), "v3": np.array([]), "simneural": np.array([]), "subID": np.array([]) } simulated data list

sim_data=[] Generate subject-level parameters

for i in range(n_subjects):

set the seed for each subject deterministically so all models are based

on the same data np.random.seed(i)

generate neural data

simneural=np.random.uniform(0, 1, size=n_trials)

Whether to include subject-specific slope, default is False

if SubSlope:

generate v0, v1, v2, v3

v0=np.exp(np.random.normal(intercept0, param_sv) + np.random.normal(a0, param_sv)

np.log(simneural) + np.random.normal(b0, param_sv)np.log(1-simneural)) v1=np.exp(np.random.normal(intercept1, param_sv) + np.random.normal(a1, param_sv)np.log(simneural) + np.random.normal(b1, param_sv)np.log(1-simneural)) v2=np.exp(np.random.normal(intercept2, param_sv) + np.random.normal(a2, param_sv)np.log(simneural) + np.random.normal(b2, param_sv)np.log(1-simneural)) v3=np.exp(np.random.normal(intercept3, param_sv) + np.random.normal(a3, param_sv)np.log(simneural) + np.random.normal(b3, param_sv)np.log(1-simneural)) else: # generate v0, v1, v2, v3 v0=np.exp(np.random.normal(intercept0, param_sv) + a0np.log(simneural) + b0np.log(1-simneural)) v1=np.exp(np.random.normal(intercept1, param_sv) + a1np.log(simneural) + b1np.log(1-simneural)) v2=np.exp(np.random.normal(intercept2, param_sv) + a2np.log(simneural) + b2np.log(1-simneural)) v3=np.exp(np.random.normal(intercept3, param_sv) + a3np.log(simneural) + b3np.log(1-simneural))

IMPORTANT: for interpretable param rec test, make sure generate params within training bounds of LAN

only keep entries in subject_data that are in bounds

v0_inb= np.where(np.logical_and(v0>= 0,v0<= 2.5)) v1_inb= np.where(np.logical_and(v1>= 0,v1<= 2.5)) v01_inb=np.intersect1d(v0_inb[0],v1_inb[0])

v2_inb= np.where(np.logical_and(v2>= 0,v2<= 2.5)) v3_inb= np.where(np.logical_and(v3>= 0,v3<= 2.5)) v23_inb=np.intersect1d(v2_inb[0],v3_inb[0])

v0123_inb=np.intersect1d(v01_inb,v23_inb)#indices of elements that are in bound for all 4 arrays

only keep inbound elements

simneural=simneural[v0123_inb] v0=v0[v0123_inb] v1=v1[v0123_inb] v2=v2[v0123_inb] v3=v3[v0123_inb]

save to subject_params

subject_params["v0"]=np.append(subject_params["v0"],v0) subject_params["v1"]=np.append(subject_params["v1"],v1) subject_params["v2"]=np.append(subject_params["v2"],v2) subject_params["v3"]=np.append(subject_params["v3"],v3) subject_params["simneural"]=np.append(subject_params["simneural"],simneural) subject_params["subID"]=np.append(subject_params["subID"],np.repeat(i,len(simneural)))

simulate RT and choices

true_values = np.column_stack( [v0,v1,v2,v3, np.repeat([[2.0, 0.0, 1e-3,0.0]], axis=0, repeats=len(simneural))] )

Get mode simulations

race4nba_v = simulator.simulator(true_values, model="race_no_bias_angle_4", n_samples=1)

Random regressor as control

rand_x = np.random.uniform(0, 1, size=len(simneural)) sim_data.append( pd.DataFrame( { "rt": race4nba_v["rts"].flatten(), "response": race4nba_v["choices"].flatten(), "x": np.log(simneural), "y": np.log(1-simneural), "rand_x": np.log(rand_x), "rand_y": np.log(1-rand_x), "subID": i } ) )

make a single dataframe of subject-wise simulated data

sim_data_concat=pd.concat(sim_data)

model_race4nba_v_true = hssm.HSSM( data=sim_data_concat, model='race_no_bias_angle_4', a=2.0, z=0.0, include=[ { "name": "v0", "prior":{"name": "Uniform", "lower": 0, "upper": 2.5}, "formula": "v0 ~ 1 + x + y + (1|subID)", "link": "log", }, { "name": "v1", "prior":{"name": "Uniform", "lower": 0, "upper": 2.5}, "formula": "v1 ~ 1 + x + y + (1|subID)", "link": "log", }, { "name": "v2", "prior":{"name": "Uniform", "lower": 0, "upper": 2.5}, "formula": "v2 ~ 1 + x + y + (1|subID)", "link": "log", }, { "name": "v3", "prior":{"name": "Uniform", "lower": 0, "upper": 2.5}, "formula": "v3 ~ 1 + x + y + (1|subID)", "link": "log", } ], )

Screenshots NA

Additional context Add any other context about the problem here, such as the print out of the model.

— Reply to this email directly, view it on GitHub https://github.com/lnccbrown/HSSM/issues/463, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEJNTY32Z3G4GE2IPZTJDA3ZHTGT5AVCNFSM6AAAAABJMCIDGOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGM2TKMZTGA3DSOI . You are receiving this because you are subscribed to this thread.Message ID: @.***>