Meet the Winner of the Pressure Overflow Challenge

Meet Francisco Noa, the winner of the Pressure Overflow Challenge!

The challenge was to develop state-of-the art machine learning algorithms to predict downhole gauge pressure in gas wells. 

Pressure measurements are important inputs for production optimisation, simulation, and forecasting. Physical pressure gauges have significant reliability issues, resulting in costly maintenance requirements, and a loss of pressure data. This is an industry wide problem, where predictive models could offer a solution.

Francisco took out the top spot in the challenge with a best in class machine learning approach. Here you can get to know him a bit more!

I take part because I am passionate about data science, and because I want to become better every day and this is the best way for me.

Tell us a little bit about how you apply data science across in your  job and as a hobby. What do you love about it?

In my work I usually apply data science to the field of renewable energies. From monitoring, detection of anomalies, consumption prediction, etc. Right now I am developing a microservice that is capable of making predictions of production and consumption.

How did you acquire your data science skills?

I would say that I have been acquiring my knowledge progressively thanks to the day to day work, the competitions that I do in my free time and the passion that I have for data science (I am always reading things about new methods or approaches to different problems)

Do you have a favourite library to work with and if so, why? 

It depends on what for, but if I had to say which is my favourite library among all the ones I know, I would say ggplot (it is a visualization library for R). It is very intuitive and powerful.

Brainstorm constantly, think creatively in the different areas, feature engineering, feature selection, target transformation, outliers detection, etc...

How did you work on the competition? Was it a daily effort? Did you do a lot of research along the way or relied on their prior knowledge?

I started by understanding the problem and all the details of the competition. Later, I made a baseline including all the variables provided raw to establish a baseline from which to improve.To improve this baseline there are several very important things:

• Previous knowledge is very important, every day you learn new things and that allows you to think in a better way when considering the resolution of a problem.
• Brainstorm constantly. think creatively in the different areas, feature engineering, feature selection, target transformation, outliers detection, etc...
• Explore the data and get to feel comfortable with it, feel that you know the data you handle and that you are able to handle it as you please quickly and accurately.
• A lot of experimentation, for example I used the 5 daily submissions we had to try different experiments. Then, according to the metrics obtained in each experiment, I made decisions. (it is very important to try the things that come to mind, although most of the time they do not work, but sometimes they do)

I worked almost daily on this challenge, the truth is that I was quite hooked.

What did you learn through the latest competition?

I have learned that it is important to have a considerably robust code that allows you to experiment with your new ideas easily, that it is very important to keep track of the things you have tried and that if I propose to myself to do a good job I can compete quite well.
In addition, technically, I have better understood the importance of concepts such as:

• Importance of data cleaning when training a model
• Reproducibility of the code to perform experiments
• Feature selection methods 
• I have also learned from the way others have approached the problem in this competition. If I had to do it again, I think I would apply some little thing that would improve the results.

Why do you take part in data science competitions?

I take part because I am passionate about data science, and because I want to become better every day and this is the best way for me.

Connect with Francisco via LinkedIn