This library is being designed with the goal to learn explicit operations over unstructured text.
The repository is under heavy development. Here be dragons.
The project does not provide a stand-alone service as of now, just clone the repository and install via pip to import it into your own code.
git clone http://github.com/semanchester/estrella
cd estrella
pip install -r requirements.txt
pip install .
Make sure the Graphene service is set up and running. Create a file "sample.txt" with this content
The chicken crossed the road , because it could .
Run this code
#!/usr/bin/env python
from estrella.main import Estrella
main = Estrella()
main.run_pipeline("extended_pipeline", location="sample.txt")
from estrella.serialize.readable.oie import simple_print
facts = main.docs.hop(link_name="facts")
print(simple_print(facts))
from estrella.model.oie import ContextLabel
# Why did the chicken cross the road?
answer = facts.filter(subject="The chicken", predicate="cross", object="the road").hop(link_name="links", constraint=lambda link: link.label == ContextLabel.Cause)
print(answer.serialize(serialize_with=simple_print))
vectors = [f.numerify() for f in facts]For documentation, refer to the source code.
In general, Pipelines are used to read text and enrich it with external information (such as word embeddings or information extraction). The default_pipeline for example, defined in the default config config/default.conf will read a plain text file and convert it our internal representation. extended_pipeline will do just that and in addition enrich the words with word2vec embeddings and extract information from the text.
Views implement operations over collections of Concepts, such as Facts and can be serialized to be interpreted by humans or further processed with other tools.