https://github.com/vbprojects/miniprojects
A collection of personal miniprojects
https://github.com/vbprojects/miniprojects
julia jupyter python
Last synced: 4 months ago
JSON representation
A collection of personal miniprojects
- Host: GitHub
- URL: https://github.com/vbprojects/miniprojects
- Owner: vbprojects
- Created: 2022-12-20T20:56:40.000Z (about 3 years ago)
- Default Branch: main
- Last Pushed: 2023-03-26T01:04:32.000Z (almost 3 years ago)
- Last Synced: 2025-03-12T13:12:07.595Z (11 months ago)
- Topics: julia, jupyter, python
- Language: Jupyter Notebook
- Homepage:
- Size: 36.6 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# miniprojects
A collection of personal miniprojects, feel free to contribute.
## adjusted sigmoid optimization
Hard to explain, might right an article about it. It's what I refer too as surrogate objective.
### TODO
- More experiments
## Hyper Ellipsoid Classifier
Train ellipsoid classifier, I think it might make sense in latent space of a model. Also, good at finding clusters at least for moons data set.
### TODO
- Implement in higher dimensions
- classifier can currently only make positive predictions, probably because I haven't made radius an optimizable parameter.
- Try with different objective functions, ie. surrogate.
## ForwardForward
Implementation of Forward Forward algorithm described in https://www.cs.toronto.edu/~hinton/FFA13.pdf
### TODO
- ~~Figure out more accurate representation of algorithm, unsure if the program is actually implementing it despite promising preliminary results.~~
Implemented, need negative images to go further
## InterestingDistributions
A collection of probability distributions that I've come across.
### TODO
- ~~Figure out distribution to use as prior for probabilistic convex/non-convex hull through Hidden Markov Model.~~
Dead end
## Edge Detection
Currently, just a sobel edge detector
### TODO
- ~~Derive emission probabilities for a pixel being an edge given various filters.~~
Dead end, not interested enough
## kaggles
Kaggle notebooks
## kernelnetworks
Feature engineering through repeated kernel method application. For example, applying rbf on a rbf kernel. Produces interesting regressors and classifiers.
### TODO
- Figure out how to more accurately select subset of points for classifier, currently using kmeans++ initialization
- Figure out how to use more than 2 applications of rbf
- Combat Runges Phenomon for extrapolation
## minvocab
Looking at classifiers that output binary labels. Each node corresponding to a 1 or 0 through a modification of the sigmoid function.
### TODO
- If assigned binary labels MNIST labels take up 4 bits. 4 bits can represent 16 different states. Is there a way to boost classifier performance through mapping labels to more than one binary label? How do you train a machine that does that?
## misc articles
Title,
## OLS_Article
Notebooks concerning the derivation and optimization of Ordinary Least Squares problems.
### TODO
- ~~Get ready to put on medium or something.~~
DONE
Finished
## repeated_inner_outer_products
Title
### TODO
- ~~Maybe some kind of generative art?~~
Not enough there
## RotLogRegr
Logistic regression where boundary is determined through rotation.
### TODO
- How do you extrapolate this to higher dimensions?
## SVM_ARTICLE
Article on SVMS
### TODO
- ~~Implement SVC from scratch~~
- From scratch with Pegasos and not from scratch with JuMP, write content for article