ResNet-style CNNs To Predict Freshwater Algae Blooms in Satellite Imagery: Mediocre Results

ResNet-style CNNs To Predict Freshwater Algae Blooms in Satellite Imagery: Mediocre Results

Although I have no domain experience with satellite imagery, I've used convolutional neural nets with aerial photography to recognize marine debris. So when I saw the DataDriven challenge 'Tick Tick Bloom' I took a glance at the dataset …

more ...

Sentiment Analysis of Mastodon Toots is Very Easy

The Mastodon API is very straightforward, as is the OpenAI API for its NLP models. I wrote a quick proof-of-concept program to do sentiment analysis of "toots.".

more ...

Re-Identifying Manta Rays

My current project is re-identifying individual manta rays (Mobula alfredi and Mobula birostris) by their distinct belly patterns … er… ventral markings.

Photo of manta ray 'Queenie' showing distinctive markings

Every night at two spots on the Big Island of Hawai’i where I live, dive boats shine bright lights that attract plankton. Most nights, the plankton in turn …

more ...

What We Talk About When We Talk About Attention

The "attention" in ML is "what you should attend to," not "alertness." In the sentence "They crossed the <???> to get to the other bank." you need to "attend to" the <???> word to disambiguate "bank". If <???> is "street" then it's "bank" as in "financial institution" (most likely). If <???> is "river" then …

more ...

How to train/test on a subset of your FastAI data

If you have a large FastAI (v2) DataLoaders and you're trying to debug something at epoch-scale (such as a custom metric), an easy way to train on a small subset of your data is:

subset_size = 100 # Or whatever
selected_items = np.random.choice(dls.train_ds.items, subset_size …
more ...

Large Language Models and the Chinese Room

The Chinese Room is a 1980 thought experiment from the philosopher John Searle. The Wikipedia summarizes the setup:

[S]uppose that artificial intelligence research has succeeded in constructing a computer that behaves as if it understands Chinese. It takes Chinese characters as input and, by following the instructions of a …

more ...

How I Failed At Kaggle Happywhale

I just deleted my intermediate data and models for Kaggle’s Happywhale competition. I did terrible, never getting much above pure random guessing. Which was frustrating, because it’s a problem in Machine Learning that I’m very interested in (and want to do more work in).

Happywhale, Sad Human …

more ...

Biggest mistake on Kaggle

Don’t join a team too quickly. Once you’re on a Kaggle team:

  • You cannot choose to leave
  • The team leader cannot choose to remove you

Unless you have a very good sense of what exactly your teammates are bringing to the competition, including their:

  • Knowledge level
  • Time commitment …
more ...

Installing Detectron2 on a Mac in CPU mode

At the risk of saying, “Yeah, it’s in the docs,” this is what I did. I think the crucial thing is installed things in the proper order, so I would advise going step-by-step:

  1. Have conda installed (brew install conda if not, I suppose)
  2. Create a conda environment with conda …
more ...

A Simple 3-Step AzureML Pipeline

Get the source code and data on Github

Illustration of pipeline graph

This demonstrates how you create a multistep AzureML pipeline using a series of PythonScriptStep objects.

In this case, the calculation is extremely trivial: predicting Iris species using scikit-learn's Gaussian Naive Bayes. This pipeline could be solved (very quickly) using this code:

import …
more ...