Leopard LinkedIn 1
12:23 09/08/2025
https://www.linkedin.com/feed/update/urn:li:activity:7359700939072598018/
Leopard p1:
A few months ago, I decided to revisit an old "bucket list" task on my list that I have been meaning to try out for some time - seeing whether we can predict the stock market using deep learning.
So I coded up some python to pull share price data for 500 UK stocks from the Interactive Brokers API, every 15 minutes, and saved it into an Azure SQL database I am running, along with social media sentiment data.
And then after a couple of months, I pulled out the results into CSV files, and wrote some deep learning code to try and detect any patterns.
It output garbage. And I realized I was trying to run before I could walk.
What I realized I needed to do first, was to generate test data where I had added an artificial pattern, and write the machine learning code to try to detect this pattern. Only by doing this could I have any confidence that the code was working. Otherwise I wouldn't be able to be sure that a failure to detect a pattern in real data was because no pattern existed, or because there was a problem in my code.
So I wrote "Leopard", a deep learning project that uses the "Keras" framework, which successfully detects a pattern in my own test data, and will have the ability to detect patterns in stock market data should any exist.
There's a lot of tech involved, from multi-dimensional numpy arrays, ml models, predictions - as well as all the basic python functions, loops, and error handling.
It's all rather interesting. And I haven't got started on the real data yet.