Author Archives: jamesdmccaffrey

There are a lot of good datasets for experimenting with machine learning classification. But there are very few datasets for ML regression experiemnts. Creating a synthetic dataset for regression is relatively easy and effective. In high-level pseudo-code: create PyTorch network … Continue reading →

One of the advantages that experienced developers have compared to early-career developers is subjective intuition. Several days ago, I implemented a program that searches for PyTorch neural network hyperparameter values (number of hidden nodes, batch size, and so on) using … Continue reading →

I’ve been using PyTorch as my neural code library of choice for the past several years. Every PyTorch example I’ve ever seen has a program structure where there’s a class that defines the network, a class that defines a Dataset, … Continue reading →

Bear with me for a moment — it’s difficult to explain the topic of this blog post. I’ll work backwards from the output of a demo program: . . . End evolution Final best soln found: [9 6 4 9 … Continue reading →

Neural networks can create powerful prediction systems. But NNs have a lot of hyperparameters that must be specified and the values of the hyperparameters have a huge effect on the effectiveness of a NN. Architecture hyperparameters include number of hidden … Continue reading →

I’ve been investigating the idea of using evolutionary optimization to find good values for the hyperparameters of a PyTorch neural network. Evolutionary optimization sets up a population of possible solutions and then: create poulation of possible solutions loop several times … Continue reading →