How to used **kwargs to increase target_accepte

Questions
1

My modle like this,

with basic_model:
    epohs = 5000

    kwargs = {'target_accept':0.85}
    step = pm.NUTS(**kwargs)

    trace = pm.sample(epohs, cores=1, step=step)
    pm.traceplot(trace)

But, I got errors,

Traceback (most recent call last):
File “trymcmc.py”, line 337, in
run()
File “trymcmc.py”, line 329, in run
step = pm.NUTS(**kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\step_methods\hmc\nuts.py”, line 15
2, in init
super().init(vars, **kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\step_methods\hmc\base_hmc.py”, lin
e 72, in init
super().init(vars, blocked=blocked, model=model, dtype=dtype, **theano_k
wargs)
File “C:\Miniconda3\lib\site-packages\pymc3\step_methods\arraystep.py”, line 2
28, in init
vars, dtype=dtype, **theano_kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\model.py”, line 723, in logp_dlogp
_function
return ValueGradFunction(self.logpt, grad_vars, extra_vars, **kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\model.py”, line 456, in init
grad = tt.grad(self._cost_joined, self._vars_joined)
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 605, in grad
grad_dict, wrt, cost_name)
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1371, in _popu
late_grad_dict
rval = [access_grad_cache(elem) for elem in wrt]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1371, in
rval = [access_grad_cache(elem) for elem in wrt]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in acces
s_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in acces
s_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in acces
s_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in acces
s_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in acces
s_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in acces
s_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in acces
s_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in acces
s_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in acces
s_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in acces
s_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in acces
s_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in acces
s_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in acces
s_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1162, in acces
s_term_cache
new_output_grads)
File “C:\Miniconda3\lib\site-packages\theano\gof\op.py”, line 711, in L_op
return self.grad(inputs, output_grads)
AttributeError: ‘FromFunctionOp’ object has no attribute ‘grad’

How can I fix it. Thanks!

2

hi, you can try the answer below:

nuts_kwargs = dict(target_accept=.90)

3

Thanks, it still does not work well.
Neither

with basic_model:

    epohs = 1500 
    trace = pm.sample(epohs, cores=1, nuts_kwargs=dict(target_accept=0.95))

or

    kwargs = dict(target_accept=.90)
    trace = pm.sample(epohs, cores=1, **kwargs)

It seems the NUTS sampler failed, it changed to elementwise auto-assignment, pls see the output.
The used sampler method does not support the target_accept keywords?

Auto-assigning NUTS sampler…
> Initializing NUTS using jitter+adapt_diag…
> Initializing NUTS failed. Falling back to elementwise auto-assignment.
Traceback (most recent call last):
File “C:\Miniconda3\lib\site-packages\pymc3\sampling.py”, line 396, in sample
progressbar=progressbar, **kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\sampling.py”, line 1515, in init_nuts
step = pm.NUTS(potential=potential, model=model, **kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\step_methods\hmc\nuts.py”, line 152, in init
super().init(vars, **kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\step_methods\hmc\base_hmc.py”, line 72, in init
super().init(vars, blocked=blocked, model=model, dtype=dtype, **theano_kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\step_methods\arraystep.py”, line 228, in init
vars, dtype=dtype, **theano_kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\model.py”, line 723, in logp_dlogp_function
return ValueGradFunction(self.logpt, grad_vars, extra_vars, **kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\model.py”, line 456, in init
grad = tt.grad(self._cost_joined, self._vars_joined)
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 605, in grad
grad_dict, wrt, cost_name)
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1371, in _populate_grad_dict
rval = [access_grad_cache(elem) for elem in wrt]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1371, in
rval = [access_grad_cache(elem) for elem in wrt]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in access_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in access_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in access_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in access_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in access_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in access_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in access_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in access_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in access_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in access_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in access_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in access_term_cache
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1021, in
output_grads = [access_grad_cache(var) for var in node.outputs]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1326, in access_grad_cache
term = access_term_cache(node)[idx]
File “C:\Miniconda3\lib\site-packages\theano\gradient.py”, line 1162, in access_term_cache
new_output_grads)
File “C:\Miniconda3\lib\site-packages\theano\gof\op.py”, line 711, in L_op
return self.grad(inputs, output_grads)
AttributeError: ‘FromFunctionOp’ object has no attribute ‘grad’

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File “c:\Users\xialang2012.vscode\extensions\ms-python.python-2019.8.30787\pythonFiles\ptvsd_launcher.py”, line 43, in
main(ptvsdArgs)
File “c:\Users\xialang2012.vscode\extensions\ms-python.python-2019.8.30787\pythonFiles\lib\python\ptvsd_main_.py”, line 432, in main
run()
File “c:\Users\xialang2012.vscode\extensions\ms-python.python-2019.8.30787\pythonFiles\lib\python\ptvsd_main_.py”, line 316, in run_file
runpy.run_path(target, run_name=‘main’)
File “C:\Miniconda3\lib\runpy.py”, line 263, in run_path
pkg_name=pkg_name, script_name=fname)
File “C:\Miniconda3\lib\runpy.py”, line 96, in _run_module_code
mod_name, mod_spec, pkg_name, script_name)
File “C:\Miniconda3\lib\runpy.py”, line 85, in _run_code
exec(code, run_globals)
File “c:\Users\xialang2012\Documents\PythonApplication1\PythonApplication1\bgc-parameters-ana\mcmc-lai.py”, line 333, in
run()
File “c:\Users\xialang2012\Documents\PythonApplication1\PythonApplication1\bgc-parameters-ana\mcmc-lai.py”, line 328, in run
trace = pm.sample(epohs, cores=1, nuts_kwargs=dict(target_accept=0.95))
File “C:\Miniconda3\lib\site-packages\pymc3\sampling.py”, line 404, in sample
step = assign_step_methods(model, step, step_kwargs=kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\sampling.py”, line 155, in assign_step_methods
return instantiate_steppers(model, steps, selected_steps, step_kwargs)
File “C:\Miniconda3\lib\site-packages\pymc3\sampling.py”, line 81, in instantiate_steppers
raise ValueError(‘Unused step method arguments: %s’ % unused_args)
ValueError: Unused step method arguments: {‘target_accept’}

4

The main error seems to related to one of the function in your model does not have gradient. What do you see if you just do pm.sample()?

5

Thanks for your help.
What I intend to do is using PYMC3 to retrieve some paramters for one biological model. The current model like this

   varNum = 5
   @as_op(itypes=[tt.dvector], otypes=[tt.dvector, tt.dvector])
    def CallOutFunc(x):
        #print(x)
        a, b = synS.GetSynthesis(x) 
        return a, b

with pm.Model() as basic_model:

    # Priors for unknown model parameters
    beta = pm.TruncatedNormal('beta', sigma=5, lower=uniMin, upper=uniMax, shape=varNum)
    sigma = pm.HalfNormal('sigma', sigma=1)

    # Expected value of outcome
    mu, observedYGpp = CallOutFunc(beta)

    # Likelihood (sampling distribution) of observations
    Y_obs = pm.Normal('Y_obs', mu=mu, sigma=sigma, observed=observedYGpp)

def run():
    with basic_model:
        epohs = 3000

        #nuts_kwargs = dict(target_accept=.90)
        #step = pm.NUTS(**nuts_kwargs)

        trace = pm.sample(epohs, cores=1)
        pm.traceplot(trace)
        print(pm.summary(trace))
        pm.save_trace(trace, resourcePath + os.sep + 'result-' + str(epohs) + '.trace')

In synS.GetSynthesis(x), the biological model was called by using the parameter x, and I got the simulation value, then I gave the value to PYMC3 to evaluate it. That the total logic. I do not know if I made some mistakes about the logic.

So far, pm.sample works well, as shown,

PANDAS_TYPES = (pd.Series, pd.DataFrame, pd.Panel)
SynthesisSiteSimulation
Auto-assigning NUTS sampler…
Initializing NUTS using jitter+adapt_diag…
Initializing NUTS failed. Falling back to elementwise auto-assignment.
Sequential sampling (2 chains in 1 job)
CompoundStep

Slice: [beta]
NUTS: [sigma]
0%| | 0/3500 [00:00<?, ?it/s]

and I could get the paramters. When I increasd the number of retrieve paramters, e.g., from 5 to 8, PYMC would ask me to increase the target_accepte. Hence, I tried to add **kwargs to make it work.

When NUTS faiked to init, the Slice was used. It seems the Slice samper does not support target_accepte keyword?

6

That’s probably exactly the reason. I am surprise that pm.summary(nuts_kwargs = dict(target_accept=.90)) doesnt work, I will see if I can reproduce the error and then raise and issue on Github.

7

Thanks. If you cannot reproduce the error, I can sent you all the codes for test.