Saving nltk drawn parse tree to image file

Question

Is there any way to save the draw image from tree.draw() to an image file programmatically? I tried looking through the documentation, but I couldn't find anything.

Guarantor

Maybe print_to_file, for CanvasFrames. – Onagraceous

Answer 1

I had exactly the same need, and looking into the source code of nltk.draw.tree I found a solution:

from nltk import Tree
from nltk.draw.util import CanvasFrame
from nltk.draw import TreeWidget

cf = CanvasFrame()
t = Tree.fromstring('(S (NP this tree) (VP (V is) (AdjP pretty)))')
tc = TreeWidget(cf.canvas(),t)
cf.add_widget(tc,10,10) # (10,10) offsets
cf.print_to_file('tree.ps')
cf.destroy()

The output file is a postscript, and you can convert it to an image file using ImageMagick on terminal:

$ convert tree.ps tree.png

I think this is a quick and dirty solution; it could be inefficient in that it displays the canvas and destroys it later (perhaps there is an option to disable display, which I couldn't find). Please let me know if there is any better way.

Masse

Nice. I think you need to use Tree.fromstring() to build the tree from a string. – Leviathan

Yes: In NLTK 3, the Tree constructor no longer accepts a tree in string form. Updated. – Belden

Could you give me a hind how to set Tree object from list like [Tree('kerb_NN', ['Dropped_VBN', Tree('provide_VB', ['to_TO', Tree('access_NN', [Tree('to_IN', [Tree('hardstanding_VBG', ['new_JJ', 'permeable_JJ'])])]), Tree('for_IN', [Tree('vehicle_NN', ['one_CD', 'domestic_JJ'])])])])] please? I'd like to draw it. How to cast it? – Swahili

Answer 2

Using the nltk.draw.tree.TreeView object to create the canvas frame automatically:

>>> from nltk.tree import Tree
>>> from nltk.draw.tree import TreeView
>>> t = Tree.fromstring('(S (NP this tree) (VP (V is) (AdjP pretty)))')
>>> TreeView(t)._cframe.print_to_file('output.ps')

Then:

>>> import os
>>> os.system('convert output.ps output.png')

[output.png]:

Illuminance

Nice! Anyway to convert a .ps file to .png in python that is agnostic of OS? – Neap

Answer 3

To add to Minjoon's answer, you can change the fonts and colours of the tree to look more like the NLTK .draw() version as follows:

tc['node_font'] = 'arial 14 bold'
tc['leaf_font'] = 'arial 14'
tc['node_color'] = '#005990'
tc['leaf_color'] = '#3F8F57'
tc['line_color'] = '#175252'

Before (left) and after (right):

Firebug

This is helpful, not only for setting the style to match the draw() version but also for showing how it can be customized in general. – Belden

Yeah, the documentation is slim and you need to look really hard through the source code to figure out what options are available. I was amazed when it actually worked. – Firebug

Answer 4

To save a given NLTK tree to an image file (OS-agnostic), I recommend the Constituent-Treelib library, which builds on benepar, spaCy and NLTK. First, install it via pip install constituent-treelib

Then, perform the following steps:

from nltk import Tree
from constituent_treelib import ConstituentTree

# Define your sentence that should be parsed and saved to a file
sentence = "At least nine tenths of the students passed."

# Rather than a raw string you can also provide an already constructed NLTK tree
sentence = Tree('S', [Tree('NP', [Tree('NP', [Tree('QP', [Tree('ADVP', [Tree('RB', ['At']), Tree('RBS', ['least'])]), Tree('CD', ['nine'])]), Tree('NNS', ['tenths'])]), Tree('PP', [Tree('IN', ['of']), Tree('NP', [Tree('DT', ['the']), Tree('NNS', ['students'])])])]), Tree('VP', [Tree('VBD', ['passed'])]), Tree('.', ['.'])])

# Define the language that should be considered with respect to the underlying benepar and spaCy models 
language = ConstituentTree.Language.English

# You can also specify the desired model for the language ("Small" is selected by default)
spacy_model_size = ConstituentTree.SpacyModelSize.Large

# Create the neccesary NLP pipeline (required to instantiate a ConstituentTree object)
nlp = ConstituentTree.create_pipeline(language, spacy_model_size) 

# In case you haven't downloaded the required benepar an spaCy models, you can tell the method to do it automatically for you
# nlp = ConstituentTree.create_pipeline(language, spacy_model_size, download_models=True) 

# Instantiate a ConstituentTree object and pass it the sentence as well as the NLP pipeline
tree = ConstituentTree(sentence, nlp)

# Now you can export the tree to a file (e.g., a PDF)  
tree.export_tree("NLTK_parse_tree.pdf", verbose=True)

>>> PDF-file successfully saved to: NLTK_parse_tree.pdf

Result...

Narine

Would be nice, but too much dependencies and can't install constituent_treelib with Python 3.11.3 – Contorted

I've updated the Constituent-Treelib [1] library. The previous language detector (fasttext) caused several strange problems and unnecessarily inflated the dependency chain. Now it's more compact and also works with the python versions: 2.7 - 3.12 (tested via github action workflow). [1]: github.com/Halvani/Constituent-Treelib – Narine