Category Archives: PyTorch

Time series regression (TSR) problems are very challenging. There are dozens of techniques — and the mere fact that there are so many techniques for TSR indicates the difficulty of these problems and that there’s no single best approach. There’s … Continue reading →

I sometimes present PyTorch training sessions at my workplace or at technical conferences. By far the biggest pain point is dealing with attendees’ PyTorch installations. So I’ve been investigating the feasibility of using the online Colab system. Briefly, Colab is … Continue reading →

PyTorch is a neural network code library. By far the biggest hurdle for beginners is installation. Note: These instructions also work for other PyTorch versions such as 1.13.1 and 2.1.0 and 2.2.1 — just different .whl files (see below). These … Continue reading →

Anaconda is a collection of software packages that contains a base Python engine plus over 650 compatible Python packages. Anaconda version 2022.10 has Python 3.9.13. You need to install Anaconda Python before you can install the PyTorch neural network code … Continue reading →

I most often use Windows OS machines but I sometimes use Mac and Linux machines. It had been several months since I had used the PyTorch neural network library on a Mac machine so one weekend I figured I’d do … Continue reading →

I was looking at the Captum library for PyTorch model interpretability. I wasn’t entirely happy with Captum because it seems somewhat over-engineered and difficult to modify. So I decided to implement the simplest possible from-scratch interpretability for a PyTorch regression … Continue reading →

I was working on a neural network project recently, and I briefly wondered if it would be feasible to multiply all weights and biases by a constant, such as 2.0. I pretty quickly realized that the idea doesn’t make much … Continue reading →

I looked at the Captum library for PyTorch binary classification model interpretability. But I wasn’t entirely happy with Captum because it seems over-engineered and difficult to modify. So I decided to implement the simplest possible from-scratch interpretability for a PyTorch … Continue reading →