31–40 of 751 posts
Something similar to HN but more advanced. Just finished the roles part.
Now finishing the automatic import of RSS feed ( and adding the appropriate tag immediatly). Then adding "remote Actions" when posting through tags
And then i'll setup something usable. Now it's used as personal bookmarking service.
Startups definitely like to deck out their offices with some of the nicer high-end designs. Even Y Combinator's home page image carousel has the likes of the Bertoia Diamond Chair and the Nelson Saucer Pendant Lamp on show! Though a lot of startups go for the replica route - I know Airbnb went with replicas in their Dublin office.
The show Silicon Valley (https://www.seenonset.com/tvshows/174/silicon-valley) is a great example of what we do.
Project: I am creating a database of all Finnish Instagram users. It's been a long and hard task to do, but it's currently looking pretty good, I have over 250k accounts that are considered Finnish and their posts data. Now to actually think what I will do with this data... More info: I am currently building a website to showcase the data, but in the mean time I have a Instagram account where I for example once a wee…
We are almost ready to release the first edition, and do some broadcasting to those who may be interested in it.
It may help those who chooses Semantic-UI for building websites.
I'm still deciding on the best ways to crunch world data to benefit all users feature-wise (and what other worldbuilding features to expand into), but I'm working on an infographic of common tropes seen across genres now (did you know that ~2% of the population has green eyes, while nearly 60% of YA protagonists have green eyes? Fun!).
I've Been working on it for 3 years now and It's getting more and more fun every day. In fact, I have rediscovered so many things on my own that I thought never existed. My biggest "Eureka" moment was when I discovered factorization via the difference of squares method on my own by using 2D geometry. I previously did not know that such a method existed nor that GNFS is based on it. Later on I discovered that a composite number is actually a perfect square less a function of a triangular number which lead me to develop a method of factorization that is rather cool but unfortunately still not good enough. I have had literally dozens of these moments where I find a certain relationship between numbers that I later on find out is some sort of rule/conjecture that was made hundreds of years ago by some mathematician!
To some of you these might be really trivial and well known facts that you learned in school. For me however, it is something I take pride in as math was by far my worst subject growing up and I actually dropped out of college because of math. The last math course I took was in high school over 10 years ago.
I have a new found love for mathematics and algebra in general and I would encourage anyone reading this to pick a problem and try working on it just for the fun of it :D
We already have something like that in our own OpenStack web dashboard, but it's a crude hack in many ways, so I'm now splitting it out into a proper microservice with a properly defined API.
If you find this useful for your own OpenStack installation, please use and fork it. I will happily accept contributions to expand the number of supported OpenStack services.
Main problems:
* Since the resolution is very low, I found it hard to use traditional CV approaches to segment individual characters and classify those (for example: is this rectangular blob a few pixels diagonal a dash symbol or just noise)
* End to end CNN approach on artificially generated dataset is approaching usefulness. Training dataset is completely artificially generated, test dataset comes from manual labeling (~2k samples, ~50 classes). This approach is almost usable (96.8% accuracy on test dataset), but it is still not acceptable performance. Next thing I'm going to try is first train the CNN on training data as best as I can (have had >99.9% during training, but that could be overfitting) and then do another training run just on a part of my test data to "fine tune" the weights from first run and use remaining test data for cross-validation.
All existing OCR engines I tried could not handle such inputs: abby, tesseract, google cloud vision.
Problem: The process of getting thoughts from your head into an organized, written draft form isn't as fast or accessible as it could be. Project: I'm building a conversational UI / bot ( https://writing.ai ) that helps people write faster. The basic idea is that it asks you a series of questions about a topic, asks follow-up questions for more detail as needed, and when it's done outputs a completed draft. You're st…
Either way, always good to have more eyes occasionally. :)