When I was a PhD student, my group were given a lab next door to our main lab. During the clear out we found a bottle of HF at the back of a cabinet. Vividly remember us crouching down to look at it all saying “nope, not touching it!”. Department technicians, understandably, weren’t willing to take it off us. In the end I think our PI got the previous lab owner to deal with it.
> ML (particularly DL) tends to outperform "classical" statistical time series forecasting when the data is (strongly) nonlinear, highly dimensional and large.
This claim about forecasting with DL comes up a lot, but I’ve seen little evidence to back it up.
Personally, I’ve never managed to have the same success others apparently have with DL time series forecasting.
It's true simply because large ANNs have a higher capacity, which is great for large, nonlinear data but less so for small datasets or simple functions.
In any case, Transformers are eating ML right now and I'm actually surprised there's no "GPT-3 for time series" yet. It's technically the same problem as language modeling (that is, multi-step prediction of numerics), however, there is only a comparably little amount of human-generated data for self-supervised learning of a time series forecasting model. Another reason might be that the expected applications and potentials of such a pre-trained model aren't as glamorous as generating language.
> It's technically the same problem as language modeling
You're thinking of modeling event sequences which is not strictly speaking the same as time series modeling.
Plenty of people do use LSTMs to model event sequences, using the hidden state of the model as a vector representation of processes current location walking a graph (i.e. a Users journey through a mobile app, or navigating following links on the web.)
Time series is different because the ticks of timed events are at consistent intervals and are also part of the problem being modeled. In general time series models have often been distinct from sequence models.
The reason there's no GPT-3 for any general sequence is the lack of data. Typically the vocabulary of events is much smaller than natural languages and the corpus of sequences much smaller.
There's a deeper issue. All language (and code and other things in the GPT/etc corpora) seem to have something in common - hierarchical, short- and long-range structure.
In contrast, there is nothing that all time series have in common. There's no way to learn generic time series knowledge that will reliably generalise to new unseen time series.
Fluidly | London, UK | ONSITE | Data Engineer, Python Engineer, Data Scientist, Full Stack Engineer
We want to solve the biggest problem facing businesses – poor cashflow. We use technology to provide real-time cashflow forecasting and automated credit control.
We recently closed a £5M funding round led by Nyca Partners [1]. We're one of Wired's hottest European startups [2] and have received numerous industry awards. We're hiring across the team, in particular in engineering and data science.
We're a friendly and enthusiastic team using GCP, AWS and Heroku. We're about to go all-in on stream processing on GCP, so we're especially interested in people with experience in streaming. In data science, we use statistical and ML models to forecast time series and solve other problems within business finance.
Fluidly | London, UK | ONSITE | Senior and Mid-Level Full Stack Engineers and Senior Data Engineer
We want to solve the biggest problem facing businesses – poor cashflow. We use technology to provide real-time cashflow forecasting and automated credit control. This year we were chosen as one of Wired's hottest European startups and have received numerous industry awards.
We've got an amazing team and great technology stack, mostly running on GCP. It's a great opportunity to scale and build an exciting data-intensive product with machine learning at its core.
>To become a chemical engineer, you first have to complete the course work for a degree in chemistry.
That is incorrect.
Look at the programme of study for a chemical engineering undergraduate degree [1]. It doesn't cover anywhere near the same content as an undergraduate chemistry degree [2].
I was typing a similar response. I did a 4-year MEng in ChemEng and there were only a few non-elective courses in organic and physical chemistry. The bulk of the content is fluid mechanics, thermodynamics, vessel design, safety, control systems, mathematics etc.
I probably over-generalized from my own experience. The only chemistry degree coursework we could avoid was a class about lab operations, basically safety/hazard management.
This turned it into a de facto 5-year degree even though they stripped most of the non-related coursework to a bare minimum; you still had two years of coursework for the engineering program after the chemistry. People that lost their appetite for chemical engineering after compressing all their chemistry into three years could switch to chemistry, which meant they spent their last year taking the filler non-technical courses that the engineers were allowed to avoid.
It was nearly correct in my case: While getting a BS in ChemE at CU Boulder, I also earned a got a minor in chemistry and biochemistry (and inorganic chemistry, but I was only allowed two minors for whatever reason).
There are huge overlaps between the different kinds of engineering - not quite enough to support a single unified core curriculum, but enough to make it obvious that there's a lot of similar math across all the disciplines, and only the emphasis varies.
The big problem with engineering isn't the specialisms, it's the name. Outsiders still think engineers are people who wear oil-covered overalls and are essentially mechanics.
Not one person in fifty understands that the job is mostly mathematical modelling of complex systems.
Those rates are for employees of the college. PhD students aren't employees.
PhD students in the UK typically receive a stipend from their research council. For 2017/18 it's £14,553 [1]. You don't pay national insurance or income tax on a stipend.
I lived in London on such a stipend for four years. It's tough.
DueDil’s mission is to be the largest source of private company data in the world. We want to be the fuel for a more informed and connected economy. This year we launched a new and improved version of our API and became a pan-European platform, covering 100M companies.
We are looking for a Python Developer to join the DueDil Labs team. DueDil Labs works closely with our most valued clients to understand and solve their business problems using private company data.
You will work with two data scientists and our product team to explore ideas, develop prototypes, and build new tools that produce actionable insights from company data. Collaborating with other specialists you will deploy our most successful projects to our web-based enterprise tool and API. You will also write data processing pipelines that onboard new datasets onto our platform.
For more info and to apply see the link on our website. If you have any questions get in touch: tom.phillips@duedil.com
