Parsing API about spacy

[honnibal]

Okay, thanks. This part was a little bit hasty. I wanted to polish things a bit more before release, but am travelling soon so bit the bullet and decided to put it in people's hands.

I think it's clear that there needs to be some sort of child node iterator, possibly for left and right subtrees, as well as a way to get the left and right arity of the node.

I think the positive/negative indexing I did here is bad, too. It just needs to be a sequence, -1 should always be "from the end".

How do you feel about these names?

n_left
n_right
lefts
rights
siblings

One trick could be that siblings iterates in alternating left/right order. If you need the left or right siblings, you can do token.head.lefts or token.head.rights. If you only need the two nearest leftmost, you can do siblings[0:4:2]

[simonhughes22]

That sounds reasonable. I think the main thing is to have a reliable way of iterating through the children and getting the parent. The rest is helpful. As child() is somewhat unreliable, I have to use the head property to get the full tree presently.

Simon Hughes | Chief Data Scientist | Dice Data Science Team | Dice Holdings, Inc.
Dice | eFinancialCareers | ClearanceJobs | Rigzone | HealthCallings | Targeted Job Fairs
M: 515.371.4258

From: honnibal <[email protected]mailto:[email protected]>
Reply-To: honnibal/spaCy <[email protected]mailto:[email protected]>
Date: Wednesday, January 28, 2015 at 10:37 PM
To: honnibal/spaCy <[email protected]mailto:[email protected]>
Cc: Simon Hughes <[email protected]mailto:[email protected]>
Subject: Re: [spaCy] Parsing API (#14)

Okay, thanks. This part was a little bit hasty. I wanted to polish things a bit more before release, but am travelling soon so bit the bullet and decided to put it in people's hands.

I think it's clear that there needs to be some sort of child node iterator, possibly for left and right subtrees, as well as a way to get the left and right arity of the node.

I think the positive/negative indexing I did here is bad, too. It just needs to be a sequence, -1 should always be "from the end".

How do you feel about these names?

n_left
n_right
lefts
rights
siblings

One trick could be that siblings iterates in alternating left/right order. If you need the left or right siblings, you can do token.head.lefts or token.head.rights. If you only need the two nearest leftmost, you can do siblings[0:4:2]

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/14#issuecomment-71968256.

[honnibal]

Yes, I'll add tests that verify:

for token in tokens:
    for child in token.children:
        child.head == token

[honnibal]

Leaving this issue open until I test and document this feature properly, but I think I've got this covered now.

[simonhughes22]

Note that the test above won't uncover the bug I found. The bug I found was that some nodes had missing children. However, all nodes correctly had their heads. For now I'm using it by reconstructing the tree from the head relation as that seems to work.

[yang]

@honnibal In the meantime, would you mind providing an example of how to traverse the parse tree (using the current API), e.g. to print the full tree? Thanks!

[honnibal]

I've updated the tests here: https://github.com/honnibal/spaCy/blob/master/tests/test_parse_navigate.py

This also gives an example of how to traverse the parse tree for now, until I have time to write the documentation.

[honnibal]

Documentation now contains some information about navigating the parse trees. I'll write a tutorial for this in the future as well.

[simonedu]

I am interesting in knowing how to traverse the parse tree. The link to the tests and the example suggested above is broken.

Thank you very much.

[rgregory1963]

@simonedu Here's the correct like I found
https://github.com/spacy-io/spaCy/blob/master/spacy/tests/parser/test_parse_navigate.py

[simonedu]

Thank you!

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.