michaelchughes

michaelchughes

michaelchughes

import numpy as np

import jax
import jax.numpy as jnp
import jax.nn

def calc_trans_mat(x_TD, r_KD, p_KK):
    ''' Compute transition matrix for each timestep

    Args

michaelchughes

import jax
import jax.numpy as jnp

if __name__ == '__main__':
    print("jax.devices()")
    print(jax.devices())

    a = jnp.asarray([[1.0, 2.0, 3.0], [4., 5., 6.]])
    b = jnp.asarray([[1.0, 2.0], [3.0, 4.0], [5., 6.]])

michaelchughes

Install notes from BDL2022f env install on 2022-11-28

• For CUDA TOOLKIT 11.3, which can be used on older devices but may not be optimal

1. set up basic conda env without any torch or jax packages, via

conda env create -f bdl_2022f.yml 

michaelchughes

michaelchughes

import numpy as np
import scipy.stats

import matplotlib.pyplot as plt

from statsmodels.distributions.empirical_distribution import ECDF

def create_transform_func_to_match_source(target_x_ND, src_x_MD, n_quantiles=1000):
    '''

michaelchughes

michaelchughes

''' VI for Poisson Normal

Model
-----
Latent variable z is drawn from a Normal prior: z ~ Normal( 40, 10)
Data y is drawn iid from a Poisson likelihood: y_n ~ Poisson(z)

Approx Posterior
----------------
Posterior on z is assumed to be Normal with unknown mean and stddev

michaelchughes