Machine Learning / Data Science notes

News

• mln.dev
• All AI News

Learning

• Virgilio - path for machine learning learning
• Awesome Machine Learning/Data Science repo -- all the links you could want and more
• Machine Learning Operations - curated list
• MIT Intro to Deep Learning course
• Intro to GeoML
• Earth Lab - online tutorials for earth data science
• YouTube playlist - ARM ML seminar series
• Devis Peressutti - developer for eo-learn

Data Science

• Coding for Data - free online Data Science textbook
• Hands-on Data Visualization -- awesome book

Javascript/browser

• Very cool overview of ml5.js & p5.js, with a sample implementation

Maps

• Vancouver tree map -- lots of stuff to steal here
• Eagle tracking
• Movebank.org -- animal tracking data

Geoscience

• Visualizing LIDAR data in Python

• Exploring geo data with Leafmap

• Digital Africa Sandbox Notebooks -- the "Real World Examples" section looks really good

Reference

• Machine Learning in Python, Step-By-Step -- from the excellent machinelearningmastery.com site

• The Incredible PyTorch

• Neural Networks and Deep Learning - online book, CC-licensed, excellentdy

• ML Cheatsheet

• Picking the best model

Books

• A list of great ML books (many freely available)

Production

• Deep Learning in Production -- a curated set of links

TinyML

• Towards Data Science article
• Microsoft Bonsai -- classifier that can run on Arduino
• Edge Impulse -- neat service for making models easily & deploying to Arduino, Pi, etc

Repos

• Machine Learning Deliberate Practice -- great intro to one person's project to get better at ML

• AWS Sagemaker examples

• Microsoft Computer Vision Recipes

GIS/Satellite imagery/etc

• Syam Kakarla - Deep learning for land cover classification of satellite imagery

  • Pretty sure these are the notebooks for that article
  • EarthPy library (import earthpy as ep)
  • RasterIO library (import rio)

• Excellent course notebooks on visualization for geoscience

Data sets

• Practice ml with small, in-memory datasets from the UC ML Repository
• Awesome satellite imagery datasets
• Awesome forests dataset

Techniques

• Long short-term memory analysis with Keras and Python
• Good overview of object detection & segmentation, wrapped up in a tutorial about Pytorch Lightning and DET
• miniGPT -- A minimal PyTorch re-implementation of the OpenAI GPT (Generative Pretrained Transformer) training

Tools

• Cookiecutter data science - the DS equivalent of "rails new"
• GBDX Notebooks - Jupyter notebooks + Digital Globe imagery
• EasyImages - Library for easy image display in Jupyter notebooks (mentioned in this article about image segmentation)
• Boilerplate code for machine learning -- excellent
  • Repo

Image labelling

• Awesome data labelling resources
• MakeSense.ai -- cool online, open-source image labeller

Automated optimization

• H2O - Automated model optimization
• TPOT

Libraries

• Folium - Python + LeafletJS for maps

• Categorical Encoding - Scikit-Learn library

• Segmentation models for PyTorch

• Wireservice -- excellent data science libraries. Like Pandas, but much simpler.

Engineering

• OPML Conference 2019 -- Usenix conference

Computing

• Intel for display, NVidia for computing

Projects

• Bird identification with Keras and CNNs
• Good overview of resources for bird identification

Ein-notebook: Emacs interface to Jupyter

Test from ein

This is a notebook explaining how I got ein working.

• Fire up a shell in the project, and activate the venv.
• Run jnb to start up Jupyter.
• Copy the URL, including the token. (Note: this part I'm still figuring out.)
• Run ein:login and give the URL, NOT including the token (but try the token if that doesn't work)
• This will fail. Try again; you should now be able to get a list of notebooks, etc.
• Click on the [Dir] link for a directory, rather than the name of a directory.
• You can click on [New Notebook], or run ein:notebooklist-new-notebook-with-name. Note that you'll need to include the directory with the name:
  • my_new_notebook will create my_new_notebook.ipynb at the root of the project.
  • notebooks/my_new_notebook will create that notebook in the notebooks directory.

Some helpful commands

• ein:notebooklist-open will show you a list of notebooks. Note: Hit on the [Dir] part, not the directory name part.
• C-c will execute a cell.
• M-return will execute a cell and go to the next cell.
• C-c C-a will insert a cell above the current one.
• C-c C-b will insert a cell below the current one.
• C-c C-t toggles the cell type (markdown to code & vice-versa). So will C-c C-u.
• C-c C-s will split a cell at point.
• C-c RET will merge a cell with the previous cell.
• C-c C-k will kill/delete a cell.
• C-x C-s will save a notebook.
• C-x C-w will rename a notebook. Note that if you want it saved in a directory, you have to include that in the name. Thus:
  • C-x C-w my_notebook will save it as my_notebook.ipynb in the root of the project.
  • C-x C-w notebooks/my_notebook will save it with the same name, but in the notebooks directory.