Policies like that are often designed to (or at least conceived/fantasized to) ferret out/prevent single-actor fraud.
Forcing someone to take two contiguous weeks out of the office means that someone else will need to be briefed for continuity and there's an enhanced chance of detecting "weirdness" that might be associated with a fraudulent scheme.
Yes, 'conspiracies' of more than one person are much harder to keep under wraps. That regulation was directly in response to the high-profile cases of rogue traders.
"As Jeremy Howard points out, even academic papers often use softmax for multi-class classification, and I too have already seen it used incorrectly in blogs and papers during my short time studying DL."
AFAIK softmax should be used with mutli-class classification and sigmoid can be used with mutli-label classification.
1) Tax laws are completely different for each European country.
2) At least for some of the countries (the once where I'm aware how taxes work) this is not the case: Even if you exchange one cryptocurrency for another cryptocurrency it's taxable. However, long-term capital gains of cryptocurrencies aren't taxable in some European countries.
In Estonia(within Europe) we have this awesome "investment account" system where you make a separate bank account and the tax system is simplified for that account.
Any money you put in, can be taken out without paying any taxes.
Once you've taken out more than you paid in, you just pay income tax and that's it.
You report your "paid in" and "paid out" sum(to/from that one account) to the tax office yearly.
Really makes the whole process simple and adaptable to any investment model.
The comment you responded to has a point - even I take WaPo-publications with a truckload of salts of grain nowadays and I don't even have a stake in US-politics.
At train time, the code supports multiple labels by sampling one of the k label at random. At test time, it only predicts the most probable label for each example.
We will add more functionalities for multi label classification in the future (predict the top k labels, etc...).
They represent the sequence as a bag of n-grams, and feed that into the classifier, rather than feeding the sequence directly. The paper basically combines variants on a few old techniques (although a few of the variants are significant and recent), but the interesting result is that they show that put together in the right way and tweaked a little, they're competitive in accuracy with state-of-the-art deep neural network models, at least on some problems, while being much faster to train. Section 2 of the paper, although pretty brief, is where this info is.
Specifically the bag of n-grams can be viewed as a very sparse vector with non-zero entries corresponding to the n-grams in the bag. As a result, n-grams not seen during training need to be ignored.
