Binomial Logistic Regression with multiple features

zsal · October 25, 2017, 4:26pm

I’m trying to run a multi feature binomial logistic regression and running into problems. I’m using
https://github.com/blei-lab/edward/blob/master/examples/bayesian_logistic_regression.py as reference and modifying number of features, D. However, modifying ‘D’ to > 1 will fail at the the ed.dot (line 74, because the matrix (400,1) doesn’t align with vector D > 1).

I’d also really appreciate if someone has intuition for the “inputs” initialization (line 73). Currently, “inputs” is set to 400 evenly spaced points from -5 to 3, but why is it set this way? How will it have to change if I increase D?

Thanks in advance for the help!

zsal · October 31, 2017, 2:20pm

Fortunately, the other example using importance weighted VI works with larger D.

github.com

blei-lab/edward/blob/master/examples/iwvi.py

#!/usr/bin/env python
"""A demo of how to develop new inference algorithms in Edward. Here
we implement importance-weighted variational inference. We test it on
logistic regression.
"""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import edward as ed
import numpy as np
import six
import tensorflow as tf

from edward.inferences import VariationalInference
from edward.models import Bernoulli, Normal, RandomVariable
from edward.util import copy
from scipy.special import expit

This file has been truncated. show original

Topic		Replies	Views
Resolve logistic regression parameters on simulated data	20	1865	April 28, 2018
Building a multinomial regression model	2	1625	July 3, 2017
Bernoulli Mixture Models on MNIST	3	1937	July 11, 2017
Bayesian Model Combination (Kim, Ghahramani 2012)	3	1363	August 13, 2017
Dirichlet process mixture model for multivariate Bernoulli mixture	0	743	April 22, 2018

Binomial Logistic Regression with multiple features

Related Topics