Liam Brannigan participated in an EGU careers workshop throughout the summer. Liam’s career has been varied. He worked as an investment adviser for some years after earning his master’s diploma in arithmetic from Edinburgh university. However in 2010, he decided to make a career shift and pursued a profession in physical oceanography.
He completed his PhD at the university of Oxford and his MSc at Bangor college. After a few post-docs, Liam was ready for greater security and decided to pursue a job in data science. This blog entry, which was first posted on Liam’s personal blog, describes his move from academics to the business world and the things he learned in the process.
I labored as a studies scientist in physical oceanography for two years before switching to the industry as a records scientist with Analytics Engines. I later spoke about this change at a panel organised by using the European Geosciences Union, and that I felt that I ought to move into greater intensity approximately the advantages of this adjustment and the training I discovered.
In April of 2018, I commenced operating at Analytics Engines after being contacted by means of a recruiter. In comparison to the length of time required to apply for academic fellowships, the process moved quickly—from a talk to two interviews to a job offer in only four days. I began two weeks later.
Acquiring a position
Being the company’s first data scientist was a perilous job, in a manner. Numerous individuals have discovered the dangers of this role: upon arrival, you discover a dearth of high-quality data, and you mostly dedicate your time to organising it instead of producing valuable insights. With Analytics Engines, I was surrounded by skilled data engineers who were knowledgeable in creating visuals and configuring databases, so the risk was minimal.
IRL Machine Learning
During my postdoc, I looked at Bayesian time series forecasting and did some machine learning. I quickly came to the conclusion that this experience would be restricted since, in the majority of companies, it is considerably more difficult to locate time series data that is consistent enough to provide insights beyond study.
For instance, we have decades’ worth of reports on food fraud in one project. but while this sort of rip-off is discovered, the perpetrators usually pass on to any other area, and there are many to select from. This suggests that although the general degree of fraud stays highly stable through the years, the precise styles of fraud vary significantly.
Accumulating and Preserving Information about Machine Learning
Using text analysis (NLP) on news story data, I was able to make some progress in this field. Text data is widely used in both the public and commercial sectors, but few climate scientists actually deal with it. The difficulty was that, as the lone data scientist, I had no more seasoned colleague to consult when I realized how clueless I was about text analysis.
Rather, I jumped into NLP Twitter and started following everyone who had anything intriguing. I began to grasp the current techniques and was doing fascinating things on my own after a few months of reading blog articles, attending lectures, and attempting tutorials. One benefit of entering the field of machine learning now is that advances happen so quickly that you can often catch up much quicker than in a scientific field that has taken decades to develop.
A Monthly postdoctoral fellow
Not only can tools move quickly, but projects do too. I may work on a new project each month, as opposed to the two or three projects I worked on during my postdoc. Furthermore, these projects may be quite varied; at one occasion, I was working on text analysis, illegal dumping of geographical patterns, and protein sequence analysis all at once. This emphasises the need of being able to quickly catch up with your partners who are domain experts while attempting to use their expertise as much as feasible.
It all comes down to output.
I believed I had finally succeeded after fitting my first models to real customer data with satisfactory outcomes. And I was not. Your models must be implemented into production in order to be useful. This implies that your model has to be packaged such that it can continue to function no matter what happens to it.
There are committed specialists who work on this full-time basis at huge firms. You learn how to do this on your own at startups. There are several obstacles in the way of this manufacturing shift. There are relatively few blog entries covering these more difficult deployment concerns, compared to the plethora of blogs that demonstrate how to fit a model in scikit-learn or PyTorch.
Every business is unique.
The fact that everything here is based on my experiences with a single organisation is an important consideration. There can be a lot greater push at other organisations to put in more overtime or take frequent trips. In general, I would suggest that compared to beginning as a postdoc at a new lab, you have less understanding about what working circumstances to anticipate when you start at a new firm.
I am , as an alternative, thrilled with the trade I’ve made overall. My capacity to efficiently manipulate the many demands of raising a small family has enabled me to paint hard and maintain a healthy work-life stability.
Furthermore, the job I’ve done at Analytics Engines has been just as intellectually engaging as my research. Naturally, the shift has prevented me from concentrating on the environmental issues that are important to me, but I have faith that the new abilities I’ve acquired will enable me to eventually make a new contribution.
Read more about Earth Science
For more blogs click here.