GroupBy

[sirinath]

Ability to groupby is very much needed.

[sirinath]

Writing to follow up on this.

[shashi]

I don't really know about the groupby you are referring to. PRs and more explanation are welcome.

[sirinath]

See:

• http://reactivex.io/documentation/operators/groupby.html
• http://pandas.pydata.org/pandas-docs/stable/generated/pandas.DataFrame.groupby.html

[JobJob]

@sirinath can you give an example in Julia of what the input and output would be. In particular would we return a Dictionary mapping from the group key(?) to the Signal with the corresponding values on it?

[sirinath]

You can have a GroupedSignal like done in RxJava's GroupedObservable.

[JobJob]

If you can give a vague example in Julia using Reactive Signals and how you would plan to use the results that will make things much easier. e.g.

s = Signal(2)
# group into even (key == true) and odd (key == false) Signals, with each Signal having the relevant values of s
odds_and_evens = GroupBy(sigval->sigval%2==0,  identity, s)

What do you want to be able to do with odds_and_evens from here?

[sirinath]

true_signal = get(odds_and_evens, true)

[JobJob]

ok, any other functions you need e.g. key_signal = keys(odds_and_evens) ?

[shashi]

why would filter not be enough? true_signal = filter(identity, odds_and_evens)

[sirinath]

There are many keys. Above is a special case where you can use filter when key is boolean.

[shashi]

filter(x->x in many_keys, signal_with_even_more_keys)?

[sirinath]

This should a Fan Out pattern where you start with 1 signal and get n signals corresponding to n keys. You have to use n filters to achieve the same. Say if n = 8000 the overhead of all the filtering maybe more greater.

The code shown above you get 1 signal which has many keys in the stream. Not one signal for each key.

[shashi]

Ok, makes sense. Thanks for clarifying!