Should we abandon LSTMs for CNNs?

Image for post
Image for post
Photo by MayoFi on Unsplash

LSTMs have always been considered the go-to algorithm for sequential data, and CNNs have always been considered as the best algorithm for image data processing. But how much truth is in this sentiment? Many papers have been written, on the use of character-level CNNs that are on par with, or even better than LSTMs or other recurrent networks!

To test this hypothesis, I decided to implement both of these algorithms to a classic case of time-series analysis: stock price prediction.

To give a fair comparison of how the two different algorithms match up in an experiment, I must give context…

A different approach to stock trading with machine learning

Image for post
Image for post
Photo by Arseny Togulev on Unsplash

Most applications of machine learning, or at least most of the content on medium, consists of training a machine-learning algorithm to predict the future price or future direction of a stock. This has proved to be ineffective, as stock data is not suitable for time series prediction methods. This is because there is no correlation between data points. This would make it so that any weights learned by the neural network would just be statistical noise within the data.

This made me think that maybe the problem could be with the scope of the solution: Maybe direct prediction based on…

Just for fun

Image for post
Image for post
Photo by Pietro Mattia on Unsplash

I am interested in the application of machine learning in different fields, where analysis of large amounts of data is necessary to come to accurate conclusions. One such field is sports analytics. Unlike other forms of gambling, the environment and the actors within it are sure to have an effect on the result, making it a problem that can be solved with machine learning.

I wanted to try to implementing deep learning in this field and so I looked online for datasets about sports. Eventually, I found a dataset on Hong Kong horse racing that contained 6000+ races. Unlike other…

With Playable AI generated Music Files

Image for post
Image for post
Photo by Rajesh Kavasseri on Unsplash

This project is a continuation of my other project, because I came up with an idea to generate even better results. Instead of using GANs, I will use LSTMs. This would make training faster and would allow for an objective evaluation function to evaluate the model’s performance.

Trained on music composed by human composers, the model is trained to use the past notes in order to predict the next note.This results in the model finding a pattern between past notes and the next note, that will allow for the generation of original music

Here is a generated extract that I…

Does it really improve results?

Image for post
Image for post
Photo by Wesley Tingey on Unsplash

Most stock trading algorithms that incorporate machine learning are based upon some form of linear regression. There are benefits and drawbacks to this method. The benefit of this is that the predicted prices of linear regression can be integrated into more complex values, that need the actual price values to function. The drawback is that for the basic “buy low, sell high” strategy, it is not directly related to predicting the direction of the price.

What would happen if we used logistic regression, or more specifically binary classification, to predict if the price will increase or decrease? Theoretically, it would…

Using the pix2pix architecture

Image for post
Image for post
Becker, Dan. (2017). Cityscapes Image Pairs Semantic Segmentation for Improving Automated Driving, Version 1. Retrieved December 3 2021 from

Semantic Segmentation is one of the key concepts within computer vision. Semantic segmentation allows computers by color-coding objects in an image by type.

How are GANs suited to perform semantic segmentation?

GANs are built upon the concept of replicating and generating original content, based on real-world content. This makes them suitable for the task of semantic segmentation on street-view images. The segmentation of the different parts allow the agent that is navigating the environment to act suitably.

Data Preprocessing:

I accessed the data from a kaggle dataset, where the streetview, and the segmented images are paired together. This means that in order…

Hands-on Tutorials

With playable audio files to listen to the generated music

Image for post
Image for post
Photo by Marius Masalar on Unsplash

Here is the github repo (ads) of this project:

GANs are highly versatile, allowing for the generation of anything that can be synthesized into images. By utilizing this feature of GANs, it is possible to generate very unorthodox content, at least from the perspective of machine learning. This article is sharing my project where I used GANs to generate baroque music, based on midi files of Bach compositions.

This is not the first time that I used GANs to generate creative content. I wrote a GAN last time that would generate art, based on famous artworks by famous artists. …

Does more data make the program more or less reliable?

Image for post
Image for post
Photo by Omar Prestwich on Unsplash

Most trading bots rely on one of two things: Machine Learning or Technical analysis. Each method has its own benefits and drawbacks. Machine Learning can find complex patterns but at the cost of computational speed and overfitting. Technical analysis can find time-tested patterns quickly, but suffers in terms of accuracy. What would happen if you combined both technical analysis and deep learning together? Would it create the best of both worlds or suffer from both of the drawbacks?


This project is not a deep dive into connecting technical analysis and machine learning, but just a simple framework so that the…

Built with Keras and Python

Image for post
Image for post
Photo by Damiano Lingauri on Unsplash

GANs are one of the most important advancements in the field of machine learning: From generating creative original content, to converting satellite images into street view. This is due to its innate ability to ‘mimic’ the data that is fed to it.

Additionally, I wanted some more experience on conditional GANs (GANs that evaluate how likely the new transformation is plausible) as they feed into much more complex and challenging problems.



In terms of data, I downloaded a large PGN file (3000+ instances) on Magnus Carlsen’s (the top chess player as of right now) games online. These files are a…

Exploring new ideas such as resistance, strengthening and architectural restructuring

Image for post
Image for post
Photo by Aaron Burden on Unsplash

LSTMs was and still is an essential part of the growth of machine learning. It allows for temporal pattern recognition, in the same way that convolutional networks allow for spatial pattern recognition. Ever since, it was proposed it has mostly remained the same. How much better could it be if we tried to change/experiment with its architecture and features?

A few months ago, I built a trainable LSTM from scratch. This will be used as the control for the experiment. I will be adding, removing, magnifying and downscaling certain features and observe their effect on convergence.

Let me explain what…

Victor Sim

Limited to 4GB of RAM.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store