harshagrawal13 / gawler-unearthed Goto Github PK

Hey there! A very heartful welcome from the team of Mine-Now. We have tried to predict mineralization in the Gawler region using Machine Learning & Data Processing. In this repository, starting from data processing to using Machine Learning, we have mentioned each and every step essential along the way.

We have found ~350K new & novel mineralized locations in the Gawler region using three different Machine Learning models. All the models that were trained attained an AUC of more than 0.97 individually and overall accuracy of ~88%, ~90% & ~91.5% respectively on the testing set.

This plot represents the distribution of mineralized & unmineralized datapoints (predictions of the first model) across the Gawler region.

This plot represents the distribution of different minerals across the Gawler region

This plot represents the distribution of different sizes of minerals across the Gawler region.

Before we dive right into the code, data & all that other daunting stuff, let me set something off the floor.

If you are in a hurry & simply want to jump right into the code, you can check the documentation (here) & the notebooks (here)

We had to start from the very beginning & learn everything from what a mineral is to running our three-step prediction algorithm over all the unexplored regions of Gawler. And just so you don't have to go through all the pain, we've curated a one-stop solution for you. You can follow along if you are

• A Machine Learning Enthusiast,

• AI Activist, Data Scientist,

• Geologist,

• Any combination of the above or

• Maybe a complete beginner in both the fields (mining + ML),

I guarantee you that you can still follow along with every aspect of this project.

Another good thing ->

• If you don't have a powerful system for running Machine Learning code, or

• If you don't have the best internet in the world,

You can STILL follow along!

We have tried to include multiple different ways to tackle a problem so if one somehow fails, you can opt the other one and can still progress. Along with this we have documented each and every type of errors & issues you might face while working.

There are three models that we have used to predict mineralization. The description of the repository structure is mentioned down below-

• You can find all the models and the required notebooks and data processing in the models folder.
• We have created a docs folder and we strongly suggest you to go through them to get a complete insight of the theory behind the magic.
• The folder helper_files contains the helper notebooks and a Markdown file explaining how and where you can use them.
• The folder plots contains all the plots we have plotted along the way.
• The folder res contains any other images or graphics required.

(p.s - you'll find a lot of Gifs along the way ;=) )

This was possible only by reading through a lot of previous submissions & forums that were made available on the Challenge's Forum page.

Coming back, we might not be able to cite all the already existing notebooks & developers but we credit them wholly & solely for their work!

Also, there is still a lot of work that could be done but I hope this repository can act as a stencil for further progress.

Now that everything is said; let's begin!