posterior_pred_check()

[DunovanK]

Hello all,

In the new How To page there is a section explaining the use of posterior_pred_check().  I have tried implementing this tool:

post_pred_check(model, samples=50, bins=100, stats={'mean': np.mean, 'median': np.median}, evals=None, plot=False, progress_bar=False)

When the script reaches this line the terminal displays "Sampling..." and eventually returns to the prompt without returning the results dictionary.  Any idea why this isn't yielding any output?

Best,

Kyle

[Thomas Wiecki]

That is odd. The same line returns me the expected result.

I don't think this will solve your problem but in hddm there is already some standard statistics for RT data included, so you could call:

hddm.utils.post_pred_check(model, progress_bar=False)

Does it take a little while or does it exit immediately?

Thomas

 

Best,

Kyle

[DunovanK]

Hi Thomas,

Sorry for the delayed response.  That resulted in the same issue.  It remains "Sampling..." for quite a long time and finally returns to the prompt. Here is my full script:  

#!/usr/bin/python

import hddm

from hddm.utils import post_pred_check

from kabuki.analyze import *

import pylab

# Load data from csv file into a NumPy structured array

data = hddm.load_csv('Group_AllPriors_SubEWMA.csv')

# Create a HDDM model multi object

model = hddm.HDDM(data, depends_on={'v':'object', 'z':'condition'}, include=('T'), bias=True, is_group_model=True)

# Create model and start MCMC sampling

model.sample(50000, burn=20000, thin=2)

# Print fitted parameters and other model statistics

model.print_stats()

# Plot posterior distributions and theoretical RT distributions

plot_posterior_nodes(nodes=[model.nodes_db.ix['z(90F)']['node'], model.nodes_db.ix['z(90H)']['node'], model.nodes_db.ix['z(70F)']['node'], model.nodes_db.ix['z(70H)']['node'], model.nodes_db.ix['z(neutral)']['node']], bins=50)

pylab.show()

model.plot_posteriors()

model.plot_posterior_predictive(savefig=True)

hddm.utils.post_pred_check(model, progress_bar=False)

#post_pred_check(model, samples=50, bins=100, stats={'mean': np.mean, 'median': np.median}, evals=None, plot=False, progress_bar=False)

Any other suggestions?

Thank you,

Kyle

[Thomas Wiecki]

Hi Kyle,

You just run that from the command line (e.g. python myscript.py)? In this case it looks like you are ignoring the output value and it gets discarded (this function doesn't plot automatically but just returns). Try adding a print before the statement.

Thomas

[DunovanK]

Hi Thomas,

Yes, I'm just running that script in the terminal.  I included the print and after sampling it returned

<class 'pandas.core.frame.DataFrame'>

MultiIndex: 1800 entries, ('wfpt(highF.face).1', 'std_ub') to ('wfpt(neutral.house).9', 'accuracy')

Data columns:

observed       1772  non-null values

credible       1800  non-null values

quantile       1772  non-null values

SEM            1772  non-null values

mahalanobis    1772  non-null values

mean           1800  non-null values

std            1800  non-null values

2.5q           1800  non-null values

25q            1800  non-null values

50q            1800  non-null values

75q            1800  non-null values

97.5q          1800  non-null values

NaN            1800  non-null values

dtypes: float64(13)

Would you recommend just entering each command in a python terminal instead of running a script?

Kyle

[Thomas Wiecki]

On Thu, Sep 20, 2012 at 2:25 PM, DunovanK <duno...@gmail.com> wrote:
> Hi Thomas,
>

> Yes, I'm just running that script in the terminal. I included the print and
> after sampling it returned
>
> <class 'pandas.core.frame.DataFrame'>
> MultiIndex: 1800 entries, ('wfpt(highF.face).1', 'std_ub') to
> ('wfpt(neutral.house).9', 'accuracy')
> Data columns:
> observed 1772 non-null values
> credible 1800 non-null values
> quantile 1772 non-null values
> SEM 1772 non-null values
> mahalanobis 1772 non-null values
> mean 1800 non-null values
> std 1800 non-null values
> 2.5q 1800 non-null values
> 25q 1800 non-null values
> 50q 1800 non-null values
> 75q 1800 non-null values
> 97.5q 1800 non-null values
> NaN 1800 non-null values
> dtypes: float64(13)
>
> Would you recommend just entering each command in a python terminal instead
> of running a script?

You would still get the same output (although IPython is very useful
in general).

This is the output if a pandas DataFrame (which is what we store the
results in) is too large. You can force it by calling .to_string() on
the dataframe. Or you can only output the variable you are interested
in, e.g. results['SEM']. For hacking on these kind of things I
definitely recommend IPython which has autocomplete and everything.

Thomas

[DunovanK]

Alright Thomas, I will try IPython.  

As always, thank you very much for your help.

Best,

Kyle