gadangadang / fys-stk4155

Machine Learning projects from the course FYS-STK4155 Applied Data Analysis and Machine Learning

Fys-Stk4155

This is the project repository for the course Fys-Stk4155. Grade:

Collaborators are Sakarias Frette, Mikkel Metzsch Jensen and William Hirst.

• Sakarias Frette is doing a master thesis in Computational Physics on Unsupervised Learning on particle collision data.
  
• William Hirst is doing a master thesis in Computational Physics on Supervised Learning on particle collision data.
  
• Mikkel Metzch Jensen is doing a master thesis in Computational Material Science on Machine learning on karegami for study of material properties
  

Project 1 explores three different linear regression models, OLS, Ridge and Lasso, and studies their accuracy approximating the famous Franke Function. Score: 100/100 points

Project 2 explores neural networks, and compares self made ones to networks from Tensorflow, as well as linear and logistic regression models on the Franke Function. Score 94/100 points

Project 3 explores using neural networks to solve PDE's and eigenvalue problems using Tensorflow infrastructure. It also compares against explicit solvers and numerical eigenvalue computation. Score 99/100 points
Extra: Uses boosting, decision trees and neural networks to look at Bias-Variance trade off
for housing data set. Score 27/30 points

To install requirements do:

pip install -r requirements

If one is using the Macbook M1 laptops, (Air or Pro), one needs to take a few more steps in order to run tensorflow. Assuming one does not have macOS 12.0+ installed, follow these steps here. If one have that version, simple follow the steps listed in the source at the bottom of the page.

First, remove current conda software, and install conda with miniforge

chmod +x ~/Downloads/Miniforge3-MacOSX-arm64.sh
sh ~/Downloads/Miniforge3-MacOSX-arm64.sh
source ~/miniforge3/bin/activate

Then, install the tensorflow dependencies with conda:

conda install -c apple tensorflow-deps

Assuming that one has previous versions of tensorflow for mac and metal for gpu, run the commands below. If not, skip this step:

# uninstall existing tensorflow-macos and tensorflow-metal
python -m pip uninstall tensorflow-macos
python -m pip uninstall tensorflow-metal
# Upgrade tensorflow-deps
conda install -c apple tensorflow-deps --force-reinstall

If one wants a specific tensorflow version, download version by specifying:

conda install -c apple tensorflow-deps==2.6.0

here using the example of version 2.6.0.

Now, all we need to do is to install the MacOS tensorflow package and the GPU runner, i.e Metal. Here it is important to download the correct Metal version, as the newest one requires the MacOS 12.0+ operating system update. Version 0.1.2 is stable.

python -m pip install tensorflow-macos
python -m pip install tensorflow-macos==0.1.2

Having done this correctly, one should be able to run Tensorflow with ease on a M1 MacBook.

Source: https://developer.apple.com/metal/tensorflow-plugin/