K Nearest Neighbors - GM Challenge

Inspiration

We were excited to take on this challenge because it is both potential and challenging at the same time. We got to work on real-life projects that can make an impact while using our machine learning knowledge.

What it does

We were trying to get the coordinates of the transformation matrix based on the given image.

How we built it

1. We tried to understand the meaning of the starter code and installed everything that was necessary.

2. After careful consideration, we decided to use K-nearest neighbors for our model because our data consists of numerical values and our data is two dimensional. K-nearest neighbors is a technique which involves taking a subset of k data points and basing an estimation of those k-data points. These are the perfect reasons to use K-nearest neighbors. We used Scikit-learn to help us perform the technique of K-nearest neighbors.

3. To begin with our process, we create two test sets. Each of our test sets will consist of 500 values. The first test set is our transformation matrices, and the second test set is the depth images produced by our transformation matrices.

4. After producing our test sets, we then flattened both our matrices and our depth images. We edited the estimator function accordingly and reshaped our matrices and images. After that, we use our model to produce our output and the error is 7.04. We improved the error compared to the starter code.

Challenges we ran into

The first challenge we ran into was setting up. However, after a while of trying and getting help, the problem get fixed. After that, we also ran into some problems while creating the training data set. We also got it figured out after some trials and errors. The biggest challenge that we were facing was finding the correct approach to build and train our model. We were aiming for neural networks at first, however, that plan got eliminated since there are so many details that we needed to consider and it was overall not the best option. After a while, we decided to use K-nearest neighbors, and found that it is the best option.

Accomplishments that we're proud of

We are proud that we got to collaborate and worked on this problem as a team. We were trying hard and didn't give up although this project can get very challenging at times. We found a way that worked at the end.

What we learned

We obtained more knowledge about machine learning while tacking on this project. We learned to be consistent and thinking together as a team to obtain a final result that worked out.

What's next for K Nearest Neighbors - GM Challenge

Since this is the very first Datathon for most of our team members, we hope that we can continue to improve our machine learning and data knowledge and take on new challenges at future Datathons.

Built With

• jupyter
• numpy
• python
• scikit-learn