I want to draw N independent (but not identically distributed) samples from a conditional Bernoulli distribution: y[i] ~ Ber(p[i]), where p[i] = logistic(sum(w .* x[i])). I am currently using the code below, but it seems overly complex. Is there a better way? Random.seed!(1) D = 5; N = 10; X = rand…

Eg (the last line is the sampling) using Random, StatsFuns Random.seed!(1) D, N = 5, 10 X = randn(N, D) w = randn(D) y = logistic.(X * w) .< rand(N) For the softmax case, you are essentially sampling from a categorical distribution, look at Distributions.Categorical .

Thanks a lot. For future reference, this is what I came up with. """ logreg_sample(W, X) Sample from a logistic regression model with C classes and D features. Inputs: `W`: C*D matrix of weights. `X`: N*D matrix of feature vectors. Outputs: `y_cat`: N-vector of integer labels from {1,...,C}. `y…

One annoying detail I came across: in flux.jl, it seems that design matrices always store the examples along the columns, so they have size D x N instead of N x D. This required the following change to Tamas’s code: function sample1(w, X) D, N = size(X) W = reshape(w, (1, D)) return (vec(Flux.si…

[image] murphyk: One annoying detail I came across: in flux.jl, it seems that design matrices always store the examples along the columns, so they have size D x N instead of N x D. This required the following change to Tamas’s code: If N is the number of observations, I think this is quite na…

Isn’t it just two different paradigms - classic data analysis (e.g. with DataFrames) usually has observations as rows, machine learning as columns. (MultivariateStats uses a bit of both, I think determined by the wind direction at the time of coding.) There are good reasons for both but it’s also so…

Yes I guessed that memory layout is why. It’s interesting that nearly all the ML textbooks stick to the convention that the design matrix X is N x D. The only exception I know of is the Gaussian Process book by Rasmussen and Williams (IIRC). It actually simplifies some of the equations (as well as t…

Maybe this helps explain it: https://docs.julialang.org/en/v1/manual/performance-tips/#Access-arrays-in-memory-order,-along-columns-1

I also found this related thread: [image] About the convention for the design matrix (row-as-observation or column-as-observation) Statistics At the risk of resucitating an old discussion… I was exploring some of the code in JuliaStats and Clustering and realised that a …

[image] mkborregaard: Isn’t it just two different paradigms - classic data analysis (e.g. with DataFrames) usually has observations as rows, machine learning as columns. (MultivariateStats uses a bit of both, I think determined by the wind direction at the time of coding.) There are good reasons…

How to sample from a logistic regression model

New to Julia

murphyk February 6, 2019, 1:15am 9

I also found this related thread:

Topic		Replies	Views
Anyone developing multinomial logistic regression? Optimization (Mathematical) proposal	19	4471	October 6, 2022
Mamba - logistic regression (bernoulli response). New to Julia question	4	833	November 19, 2018
Logistic regression in flux New to Julia flux	1	2098	April 8, 2020
Simulate data for logistic regression General Usage	3	1755	September 5, 2018
Generalization of the Turing Bayesian logistic regression example Statistics turing	1	535	December 6, 2022

How to sample from a logistic regression model

Related topics