gibbs_program

Project：Increasing Pharmaceutical Coating Rate

Description ：

The project is based on pharmaceutical data aimed to fix the problem that the number of regression parameters is larger than the observation number. Because the data is real Pharmaceutical production data, different medicines have different batches. Each batch is response variable, each batch has different production records. Actually, we can assume that some of different batches maybe belong to same class with some same attributes. Therefore, we established Bayesian Collaborative Model based on Gibbs sampling algorithm which could get the Estimation of parameters , , , , Simultaneously.

Parameter Description:

- Variable selection with 0 or 1 = 1 means != 0

- model parameter of each class which determinate beta

- sampling from

- which class the batch belongs to

- the probability the batch belongs to each class

, - Hyperparameter

Update iteration order: , , , , , ,

The Algorithm Step :

step 1:

Get the Bayesian posterior probability derivation-- the Algorithm basis

The folder Bayesian posterior probability derivation include the entire derivation process

step 2:

Initialize the generated simulation data

step 3:

Define Gibbs update iterator functions in the class

step 4:

Select -class number based on

In addition :

gibbs_with_same_obs.py --- under each batch having same observation number condition

gibbs_with_different_obsnumber.py ---under each batch having different observation number condition