problem when using empty dseries to create a dseries

tseries = dseries();
a = tseries(randn(3,1)) % doesn't work

@houtanb I do not remember why I allowed for empty dseries... But we could adopt the same behavior than for the dates object.

@stepan-a the idea was to provide similar functionality between dates and dseries

Empty dseries are useful for concatenating dseries in a loop or, as you do with dates initializing some member of the dseries then using that to initialize new dseries

@houtanb I understand that empty objects are useful for concatenation and initialization of loops, but I do not get what we can do with an empty dseriesobject having an initial date or a range of dates. In your example we would want to create a new dseries object with default initial date (1) and default name for the variable ('''Variable_1''')... This is not very useful. If we do (as we would with dates):

data = dseries(dates('1950Q1'))

and

a = data(rand(3,1))

we would create a new object with 1950Q1 as the initial date. But if we do:

a = data('mygdpdata.csv')

and if the first column of the CSV file has dates, how do we define the initial date then ? Also having an initial date or a range of dates in an empty dseriesobject is not useless for concatenation and loops.

I am not convinced by the idea of having similar behavior between dates and dserieson this matter, because I do not see any practical advantage... So I wonder if we should not remove the possibility of empty dseriesobjects with dates.

@stepan-a it's precisely useful for when you want a different default for a new dseries and it's useful for concatenating in loops. Aren't those reason enough?

In the example above, I like the idea that if the csv file has dates in it, the date in data would take precedence. Otherwise, the person should just use dseries.

In the end, however, I created this ticket when creating the slides for the summer school and I noted that you had created this functionality but that it didn't work. In the end, the decision is to either remove it or add this functionality. I would push to add because concatenation of dseries in loops is useful as is initialing dseries with a different default date.

if you concatenate the empty object data, defined in my previous comment, with say a=dseries(rand(3,1),'1950Q1'), you loose the information about the initial date in data. So I do not see how this could be useful. I just proposed to remove empty dseriesobjects with dates, not all the empty dseriesobjects.

I understand your idea now... This would just be a way to change the default initial date (which is 1). But

1. In this case we do not need to consider range of dates in empty objects (because there is no default for the ranges),
2. The initial date defined in the csv file should prevail (this is how it works with the default initial date),
3. The implementation is not obvious...

An empty dseries is useful for doing the following, which we should support

a=dseries()
for i=1:100
    a=[a dseries(i*2*ones(3,1))]
end

I think the initial date in the csv should prevail if using dseries. The idea of having an empty dseries object with a different inital date would be to allow the user to easily change the default initial date of any dseries created with this object, either in the code or via a .csv.

You mean a csv files without dates, correct ?

Or even one with dates, but that is up to you if you do not want to support that.

Done by closing the duplicate #21 (closed).

closed

mentioned in commit 67077fe8

mentioned in commit 62d86161

mentioned in commit 870a3a68