Justin Mackie

I extract value from data.

Recent posts

Oct 4, 2022
Transfer Learning with a One-Dimensional SignalThis project predicts pump cavitation using 16,000 Hz accelerometer data (example below). This data can be automatically categorized by a model. Transfer learning reuses a model built for an old task as the starting point for a new task. The pre-trained VGG16 Convolutional Neural Network (CNN) model was the starting point. It categorizes images. The original VGG16 was fine-tuned to categorize new images. The new images were derived from an accelerometer signal from the pump.…
Sep 28, 2022
Fourier and Bispectral Analysis of SignalsThis is a math overview of the fourier and bispectrum signal processing techniques. Computational challenges and use cases are discussed. Please see below: Fourier and Bispectral Analysis of Signals …
Nov 24, 2021
YouTube Dislikes - Sentiment AnalysisOn November 10, 2021, YouTube announced that video dislike 👎 counts would be hidden from the public. The stated benefit is protecting creators (especially smaller ones) from dislike attacks and harassment. The number of dislikes is only available in YouTube Studio to content creators. See the sentiment analysis below. It uses public comments scraped from the web and the DistillBERT language model. The YouTube Dislike Button Remains, with Number of Dislikes Hidden from Public …
Feb 5, 2020
Auto Price Prediction from Scratch!This is a four-part article series on auto price prediction from scratch. Part 1: Overview Build a dataset from the web. Prepare and model the dataset. Part 2: Data Collection and Cleaning Extract data with Scrapy Part 3: Feature Engineering Building Features for Data Modelling Part 4: Algorithms and Experiments Modelling our hard-won data …
Oct 3, 2019
Futures Trade SelectionRank Seasonal Oil Trades and Backtest Profit/Loss Copyright © 2019 Justin Mackie. All Rights Reserved. No one may distribute or create derivative works from my work without my written permission. You are prohibited from using the code for commercial and/or business purposes. Background: Suppose we want to predict oil prices several months ahead. Oil futures prices are daily time series data. Each business day, there is one settlement price for the commodity.…
Sep 12, 2019
Confusion Matrix VisualizationEasy-to-read multiclass confusion matrix. Both “normalized” percentages and counts are shown. The matrix values are color-coded with a heatmap. Illustration and code here: Confusion Matrix Visualization …
May 9, 2019
Bot AttacksThe internet is a cesspool of spy bots. Bots can attack an internet server hundreds of times a day. This Jupyter notebook, written in Python, visualizes bot attacks. It uses my anonymized server log and IP address geolocation. The illustration and code are here: bot-attacks: Server Bot Attack Visualization …