I have these variables expense_chance = np.random.binomial(1, 0.1, size=1000) expense_size = np.random.exponential(200, size=1000) expenses = expense_size * expense_chance df_data = pd.DataFrame({'expenses': expenses}) Then [image] and the final distribution of the data looks like this: [image…

I might suggest something like what I have below, in which you tackle the 2 components separately. First, you figure out what proportion of expenses are non-zero. Second, you figure out how to describe the non-zero expenses. I don’t know how much faith you have that this particular set of assumpt…

Sampling Bernoulli + Exponential cannot reproduce original distribution

Questions

tadas-subonis December 14, 2020, 1:48pm 9

Thanks. I’ll have that in mind. I could replace that with some positive-only distribution. Probably even Beta would work here.

Topic		Replies	Views
Mixture with custom distribution class Questions	1	606	October 25, 2018
Bernoulli samples all the same? Questions	1	2449	May 29, 2018
Modelling a mixture of binomial and gamma Questions	2	982	February 12, 2019
Sampling from a learned mixture of zeros and lognormal Questions	8	1475	August 8, 2019
The zero-inflated exponential Questions	5	1549	January 18, 2019

Sampling Bernoulli + Exponential cannot reproduce original distribution

Related topics