Is there any example on how to use/create an empirical distribution?

Tutorials
1

I googled empirical distribution examples however found no code examples

my code is :

def bayesian_volume_model(tick_data_series):
	p_sell_lambda=sample("lambda", Empirical())	
	for each in range(tick_data_series.__len__()):
		sample("obs_s_{}".format(each),Poisson(p_sell_lambda),obs=tick_data_series[each])	
	pass

def bayesian_volume_guide(tick_data_series):	
	sample("lambda",Empirical())
	pass

it raises exception as:

  File "D:\bin\miniconda3_x64\lib\site-packages\pyro\distributions\empirical.py", line 109, in sample
    idxs = self._categorical.sample(sample_shape=sample_shape)
AttributeError: 'NoneType' object has no attribute 'sample'

EDIT added backtick formatting

2

The Empirical distribution is mostly used internally to store weighted samples from the posterior distribution, e.g. from MCMC or importance sampling. I don’t think you would want to use this in your model directly. It will help if you explained what is it that you are looking to do - why does p_sell_lambda have to be from an “empirical” distribution?

3

the data of stock sell/buy order follows poisson distribution, whose lambda is frequently changing. here I use empirical for the lambda parameter.

the continuous incoming data refreshes the posterior lambda parameter.

4

The Empirical distribution shouldn’t be used here. You could place a gamma prior on lambda instead. If you meant that you are looking to model how lambda changes over time, I suppose you will need to build that into your model.