*(Sandipan Dey, **11 August 2016)*

- In this article, the Poisson-Gamma Conjugate Priors will be used to compute the Posterior Probabilities of the number of
*customer arrivals*in a retail shop every*10 minutes time window*(can be modeled by a*Poisson process*). - First, a Γ(α,β) prior is chosen with α=4,β=0.4 (consistent with our belief that the mean number of customers arrived in the store in a 10-min time window is 10, with a standard deviation of 5) to model the unknown mean customer arrival rate λ variable, so that λ∼Γ(4,0.4).
- Then a few trials of a random experiment simulating the customer arrival process are conducted to collect the data and update the prior belief about λ from the likelihood, which can be modeled as
*i.i.d. Poisson*random variables, Yi∼Pois(λ),λ∼Γ(α,β),∀i. - The posterior probability distribution is also a
*Gamma*distribution as shown in the figure below from the videos of professor Herbert Lee.

- Then the
*recursive Bayesian updates*and the*prior*and*posterior hyper-parameters*and the means are updated as and when a new datapoint is received. Also, the*frequentist’s MLE*and*95% confidence interval*are computed, along with the Bayesian*95% credible interval*. - The following animation shows the results of simulation of customer arrivals in 20 such time intervals (each of 10 mins), starting with the prior Γ(4,0.4).
- The left bottom barplot visualizes simulated # customers arrived in every 10 mins window.
- Every time a new datapoint is received (# customers arrived at the shop in the next 10 mins window), the prior belief is updated.
- The right bottom table represents the summary statistics. Prior and Posterior means respectively correspond to the previous and updated beliefs about the #customers arrived at the shop in a 10 mins time window.
- The next animation shows the same results starting with a
*vague prior*Γ(ϵ,ϵ).

Advertisements