Datarion - The Forecasting Factory

Koti
Suomi
Hyvinkää
Datarion - The Forecasting Factory

Datarion is an elite AI-company located in Hyvinkää, Finland. It offers expertise in machine learning and computational algorithm design since 2015.

Datarion has been at the forefront of data science since 2007, offering expertise in machine learning and algorithm design. The company combines machine learning and statistical methods with software engineering principles to build automated near real-time streaming analytics solutions where models are optimized with modern metaheuristics. Datarion provides the art of data science consulting and unique professional training in the field.

31/12/2020

Happy New Year from Datarion! Attached picture is from Ylläs, Finland, where we just did 30 km of skiing and other 30 km of hiking in a pretty cold -25 C temperature. It is again good to go back a little and look at the big picture, what happened in 2020. Guess this kind of time increases the polarization in the world between those who can adapt and those for whom it is harder. During the Corona episode, technology helped customers and corporations to stay in touch despite the physical social distancing. In Finland, we have pretty strong network infrastructure and good remote working capabilities. 5G networks spread pretty quick this year, even if the current version still uses only 4G frequencies and we have to wait until next year for the full-scale 5G.

For us data workers, it was easy and natural to work mainly remotely, but older more traditional jobs had to start learning faster how to use modern technology efficiently. Many employees saw for first time apps like Discord, Meet, Zoom and Teams and how the old doorway 'spontaneous' discussions now required a bit more energy and few mouse clicks. Better webcams were sold out from the IT stores, and after some very nice product launches from semiconductor manufacturers, they still couldn't deliver enough goods. We saw, for example, new 5 nm TSMC transistor process, AMD Ryzen 5000 series CPUs and NVidia RTX 3000 series GPUs.

Financially Datarion's Corona-year was profitable and turnover increased nicely from 2019. Our main consulting customers came from telecom and food industries. Next year looks promising as well. There are still many companies which are data rich, but poor in information, and that is something Datarion can help with. In our R&D, we will next year focus on real-time adaptive learning models, that take in new data as streams and optimize for a moving target. No more static model training every now and then!

To mention some other data highlights of the year:

Berkeley announced their free RISC-V processor in 2010 and it has been growing more adaption this year. For example Seagate announced their first RISC-V cores and plans for over 50 Tb hard-drives before 2026. Open and free architecture brings trust and true security and cryptography benefits.

Apple, Microsoft and others revealed their ARM CPU-plans meaning harder times for the traditional x86 companies such as Intel and AMD. Raspberry Pi 4 with 8 GB RAM and 400 (keyboard-model) entered the embedded markets.

PCI Express 5 doubles soon the bit transfer rates from 16 GT/s of PCIe4 to 32 GT/s. With x16 link that means approximately 128 GBps aggregated bandwidth for new applications (AI/ML, SSD, 40G Ethernet).

Jedec announced official DDR5 memory standard specification with ratings between 4800 and 6400 Mbps and 128 GB chips. The big change there is that of similar LPDDR4 and GDDR6, where single DIMMs are are being split to 2 channels; instead of 64-bit data channel in each DIMM, there will be 2 independent 32-bit data channels.

Lots of discussion about finally dumping the 32-bit software. After all we have had microprocessors with 64-bit word size since 2001. Datarion's first touch to 64-bit server machines (HP Itanium) was within a Nokia project in early 2005. About that same time, home users enjoyed AMD Athlon XP CPUs and 64-bit version of Windows XP.

That’s it for now. Stay safe and see you in 2021 again.

25/08/2020

Datarion was found 24th of August 2015, so yesterday it was time for a small 5y party with good food.

24/07/2020

Datarion is on hiking holidays and back to business on Thursday. Greetings from Norway.

27/04/2020

Earlier this month Datarion joined volunteer Covid-19 protein folding programs, that are used to predict the structure of proteins important to the disease as well as to produce new, stable mini-proteins to be used as potential therapeutics and diagnostics.

We recently changed our computational architecture from Intel to AMD Zen 2. Here is picture where our assistant was helping to decide between Intel and AMD, even she seemed to enjoy the box more than the contents.

We now donate computation time from our heaviest Zen 2 modeling machine that can handle approximately 750.000 Million Instructions Per Second (MIPS). Those interested in topic could read, e.g. https://foldingathome.org/covid19/.

23/03/2020

[v0.02] Added UI enhancements and more tooltip statistics to COVID-19 forecasting tool @ https://www.datarion.com/corona/

19/03/2020

https://www.datarion.fi/corona/ is now open. It is a small webapp that forecasts the Corona infection rates at Finnish health care districts for 7 days since the previous day. Data and models are updated every morning 10am (UTC+2 EET). Hope it helps some to prepare if the rates start to get higher.

15/03/2020

I remember the story from my mother when the Hong Kong virus was spreading in Finland in the end of 60s. Everyone from their family got it and my grandfather was so sick he was not able to move at all. The virus infected worldwide some billion people and killed approximately 1 million. [1]

Exponential growth is sometimes hard to understand as a mathematical construct. It might look a bit linear at first, and those couple of critical starting weeks are not being acted on as the numbers look still somewhat small. My old school friend Andon was recently reminding me about the old exponential mathematical joke/tale that includes the inventor of chessboard, who sold the board so that the first chessboard cell had to include 1 grain, next one 2, then 4, 8, 16, and so on, ending up to 2^64 = 18.446.744.073.709.551.616 grains. [2]

There has been lately so many postings about COVID-19 data that I decided to make few plots about the situation in Helsinki, Finland too. In sales, we often do analysis on growth and decay of product launch - customers quickly want to try out new products and we need to inform factories few weeks ahead how much to produce. After the peak sales, we need to estimate the decay effect and what might be the final base sales level thanks to the faithful re-buyers.

My COVID-19 case data is from Helsingin Sanomat newspaper and I only focus to the numbers within our main health care district HUS, as the people density is highest here, our main airport is here, and the exponential growth models apply best within limited regions - e.g. Lapland cases have nothing to do with the spreading model of our metropolitan area. In addition, Datarion office is in the very center of Helsinki so it might be good to have some plan B for April/May if it really hits the fan and people don't take it seriously here.

When other countries have closed their borders and schools, measure fever at airports etc, we still haven't done anything 'big' in the capital so far. People returning from epidemic areas have got some SMS greeting that please work from home for 14 days instead of a straight quarantine for him/her and all the close contacts.

This kind of epidemiological data has lots of uncertainty from collecting the data in individual level to the actual phenomena which is hard to model when the disease can't spread freely, and finally we debate on how some key figures are even being calculated and whether the reproduction number is at this time 2.3, 3.3 etc. [3, 4, 5] Researchers dealing with this sensitive data might not make it to news, but we hear part of the story second-hand from some directors or politicians. From the disease-based key data figures, we are usually most afraid about the mortality rate. COVID-19 phenomenom is pretty new and some sources state mortality average time of 17.3 days from the onset. Based on recent Lancet-papers, the mortality rate might be as high as 20 % [6]. Inside there are naturally major demographic factors to consider, such as in Italy, where the average mortality age was 80.3 years and 25.8 % were women.

HUS informed that they won't test all the people anymore after 14th of March when there was still clear exponential growth starting. Probably the best we can do is make some growth models now with the current limited data. I present here one infection forecast for the rest of March. For general clearance, I don't go to the log scale or percentages, but just keep the cold daily infection numbers. My estimate is that if the current growth stays stable, we will have some 10 000+ total infected people in the end the month and it reaches the speed of 3000+ new cases every day. I won't publish here estimates for as far as April/May, as the spread might not be so free anymore then (I hope it isn't), but times might get interesting in the end of May..

Remember the given instructions if you consider infection yourself. Wash hands with soap as it dissolves virus' fat membrane making it collapse like a house of cards. And one roll of toilet paper per week should be statistically enough for your royal highness - don't overdo it - especially if you live in some relaxed area with close to 0 infections.

- Datarion

[1] https://en.wikipedia.org/wiki/1968_flu_pandemic
[2] https://en.wikipedia.org/wiki/Wheat_and_chessboard_problem
[3] https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(20)30567-5/fulltext
[4] https://www.thelancet.com/journals/laninf/article/PIIS1473-3099(20)30129-8/fulltext
[5]https://www.ecdc.europa.eu/sites/default/files/documents/RRA-sixth-update-Outbreak-of-novel-coronavirus-disease-2019-COVID-19.pdf
[6] https://www.thelancet.com/journals/laninf/article/PIIS1473-3099(20)30195-X/fulltext

25/07/2019

Datarion wishes Meripäiväjuoksu people good kilometers for today!

18/07/2019

Last week Datarion spent quality time at useR! 2019 conference in Toulouse, France. Good talks for example on R core development, uncertainty measurements and current state of different automated machine learning solutions. The conference's gala dinner took place at scientific discovery center Cité de l'Espace, which also offered Ariane 5 rocket's mock-up.

27/06/2019

Datarion compensates for our computation and network bandwidth electricity consumption by renting its first 315 Watt solar panel from the roof of Helsinki convention centre.

25/01/2019

Fine dining with customers

Osoite

Senaattorintie 6
Hyvinkää
05820

Nettisivu

https://www.datarion.com/, https://www.forecastingfactory.com/

Hälytykset

Tiedä ensimmäisenä ja anna meille oikeus lähettää sinulle sähköpostitse uutisia ja promootioita Datarion - The Forecasting Factory :ltä. Sähköpostiosoitettasi ei käytetä muihin tarkoituksiin, ja voit perua milloin tahansa.

Ota Yhteyttä Yritys

Lähetä viesti Datarion - The Forecasting Factory :lle:

Oikopolut

Jaa

Kategoria

Tietokoneyritys