Algorithm Notebooks

Various Jupyter notebooks for learning new algorithms. A majority of the ML implementations use MNIST.

Any applicable credits are within the notebooks, scripts, and modules. Includes:

• Generative Adversarial Networks
  • Deep Convolutional
  • Conditional Wasserstein with Gradient Penalty
  • Inversion
• Gaussian Process Pregression
  • Carbon dioxide data set
  • General tutorial with derivations
  • xsinx using scikit learn
  • xsinx using MCMC
• JWST
  • Making source cutouts from a Webb public release observation
  • Analyze source cutouts using PCA, KMeans, and KDE
• Miscellaneous
  • Self organizing maps
  • Symbolic regression with Newtonian gravity
  • Stochastic gradient descent on linear regression
  • Principal component analysis on double gaussians
  • Kernel density estimation
• Tensorflow
  • ResNet classification
  • Time series regression
• Time Series Analysis
  • Dimensionality Reduction
    • Functional principal component analysis on temperature data
    • Functional principal component analysis on height data
    • Principal component analysis on signals with oscillating amplitudes
  • Forecasting
    • ARIMA on airline passengers dataset
    • Exponential Smoothing on airline passengers dataset
    • Comparing ARIMA and exponential smoothing
  • Signal Processing
    • Fast Fourier Transform
    • Lomb Scargle Periodograms
    • Phase folding
• UNET
  • Training and evaluating

Installations

There are three environments: one purely conda (environment_conda.yml; only pip libraries are tensorflow-macos, tensorflow-metal and minisom since they do not have conda support), one purely pip (requirements.txt; only conda library needed is scikit-fda due to incompatibility in pip), and one mix (environment.yml; preferred). All notebooks were tested in the preferred environment using a 2021 Mac M1 chip.

# create preferred env
conda env create -f environment.yml

# create conda env
conda env create -f environment_conda.yml

# create pip env
conda create -n home_pip python=3.11 pip
conda activate home_pip
pip install -r requirements.txt
conda install -c conda-forge scikit-fda

Comparing models

• Plot data (train and test) vs model (train and test)
• Residuals plot and histogram
• Plot y_test vs y_pred
• MSE, MAE, and Pearson correlation