Making it possible to use arbitrary vertex names

[iosonofabio]

@adriangb

If you want to support all combinations of sequences, labels and integer indices that’s now an exponentially growing combination of methods.
But at least the sequence vs. single item methods can be easily be done in Python (i.e. have a method that takes a single item, wraps it in a list and calls the method that takes a sequence).

That’s up to us to decide. I don’t think we need to have functions for a single vertex if there is an equivalent function that takes a sequence of vertices. Instead, I would improve the docstring and tutorials to explain our users how things work.

As of combinations of labels/integers, again I don’t think that’s useful to support. We should have a vision: either labels are first-class citizens or they are not (like now). If they are, the default is to take the sequence to mean labels: we should pass it through the fast-lookup dict to obtain the corresponding list of integers for the C layer. If the lookup fails and the input sequence contains integers, we can opt to make a second try and send it straight down to the C layer before failing. That would give some back compatibility without sacrificing the honorary position of labels from this point on.

[adriangb]

@iosonofabio

That’s up to us to decide. I don’t think we need to have functions for a single vertex if there is an equivalent function that takes a sequence of vertices. Instead, I would improve the docstring and tutorials to explain our users how things work.

As of combinations of labels/integers, again I don’t think that’s useful to support. We should have a vision: either labels are first-class citizens or they are not (like now). If they are, the default is to take the sequence to mean labels: we should pass it through the fast-lookup dict to obtain the corresponding list of integers for the C layer. If the lookup fails and the input sequence contains integers, we can opt to make a second try and send it straight down to the C layer before failing. That would give some back compatibility without sacrificing the honorary position of labels from this point on.

Yup this all sounds very reasonable. Overall I’m +1 for only accepting sequences (at least at first, helper methods can always be added later) and for only accepting labels.

[iosonofabio]

Perhaps this is useful?

GitHub - atom-moyer/getpy: A Vectorized Python Dict/Set

We would need to vectorize __hash__ and feed the hash into this.

[jboynyc]

@vtraag

@jboynyc

• As for examples for indexing tabular data in Python-land, there’s also datatable , which aims to have a nicer API than pandas .

Again, thanks for the pointer! It might be interesting to explore that further as well indeed. I do see that they actually mix integer indices and non-integer labels, but it seems difficult to use integers as labels. But I’ll have to explore it further. Did you have any experience with this package that you found particularly easy or useful?

No, it wasn’t my own experience that brought it to mind, as I haven’t done much with it yet. It was just the aim stated in the datatable documentation of improving on how indexing works in pandas.