The OxAPI Python library provides simplified access to the OxAPI from applications written in the Python language.
See the OxAPI documentation . Access to repo doc: http://github-oxapi-python-doc.s3-website.eu-central-1.amazonaws.com
You don't need this source code unless you want to modify the package. If you just want to use the package, just run:
pip install --upgrade oxapi-pythonInstall from source with:
python setup.py install├── oxapi
│ ├── abstract
│ │ └── api.py # Non-instantiable, super classes for API calls
│ ├── nlp
│ │ ├── classification.py # NLP Classification package
│ │ ├── completion.py # NLP Completion package
│ │ ├── encoding.py # NLP Encoding package
│ │ ├── pipeline.py # NLP Pipeline package
│ │ └── transformation.py # NLP Transformation package
│ ├── utils.py # General utilities
│ ├── asynch.py # package for asynchronous API calls
│ └── error.py # Custom exceptions module
├── tests # Tests
└── docs_src # Documentation source files
The library needs to be configured with your account's secret key. Either set it as the OXAPI_KEY environment variable before using the library:
export OXAPI_KEY='sk-...'Or set oxapi.api_key to its value:
import oxapi
oxapi.api_key = "sk-..."from oxapi.nlp.completion import Completion
# Performing API call
completion = Completion.create(
model="gpt-neo-2-7b",
prompt="My name is Tim.\nSentiment: Neutral\nIt is such a lovely day.\nSentiment: Positive\nAlthough I am in a bad mood\nSentiment:",
max_length=2,
do_sample=False,
eos_words=["\n"]
)
# Fetching result
res = completion.format_result(result_format="str")
print(completion.result)Output:
{'results': ['Neutral\n']}
from oxapi.nlp.classification import Classification
# Performing API call
classification = Classification.create(
model="dialog-content-filter",
texts=["I want to kill myself.", "I want to kill myself.<sep>You should not do that!", "I want to kill myself.<sep>Do it!"]
)
# Fetching result
res = classification.format_result(result_format="pd")
print(res)Output:
text label confidence_score
0 I want to kill myself. unsafe 0.9772329101403038
1 I want to kill myself.<sep>You should not do t... safe 0.9736578740966625
2 I want to kill myself.<sep>Do it! unsafe 0.9266854663680397
from oxapi.nlp.encoding import Encoding
# Performing API call
encoding = Encoding.create(
model="mpnet-base-v2",
texts=["Hello", "How are you?"]
)
# Fetching result
print(encoding.result)Output:
{'results': [[
-0.017791748046875,
-2.980232238769531e-07,
-0.022003173828125,
0.02105712890625,
-0.06695556640625,
-0.02435302734375,
-0.0174713134765625,
...
-0.0011529922485351562]]
}
from oxapi.nlp.transformation import Transformation
# Performing API call
transformation = Transformation.create(
model="punctuation-imputation",
texts=["hello my name is tim i just came back from nyc how are you doing"]
)
# Fetching result
print(transformation.result)Output:
{'results': ['Hello my name is Tim. I just came back from NYC. How are you doing?']}
from oxapi.nlp.pipeline import Pipeline
# Performing API call
pipeline = Pipeline.create(
model="en-core-web-lg",
texts=["Hi there!"]
)
# Fetching result
print(pipeline.result)Output:
{'results': [{'text': 'Hi there!',
'ents': [],
'sents': [{'start': 0, 'end': 9}],
'tokens': [{'id': 0,
'start': 0,
'end': 2,
'tag': 'UH',
'pos': 'INTJ',
'morph': '',
'lemma': 'hi',
'dep': 'ROOT',
'head': 0},
{'id': 1,
'start': 3,
'end': 8,
'tag': 'RB',
'pos': 'ADV',
'morph': 'PronType=Dem',
'lemma': 'there',
'dep': 'advmod',
'head': 0},
{'id': 2,
'start': 8,
'end': 9,
'tag': '.',
'pos': 'PUNCT',
'morph': 'PunctType=Peri',
'lemma': '!',
'dep': 'punct',
'head': 0}],
'sents_text': ['Hi there!']}]
}
With oxapi-python package is possible to make calls to OxAPI in parallel. The AsyncCallPipe class takes as input a list of API calls each set through the prepare function to be executed by the pipeline.
from oxapi.asynch import AsyncCallPipe
from oxapi.nlp.completion import Completion
from oxapi.nlp.classification import Classification
from oxapi.nlp.transformation import Transformation
from oxapi.nlp.pipeline import Pipeline
# Set up API calls
cl = Classification.prepare(model="dialog-content-filter", texts=["I want to kill myself."])
cm = Completion.prepare(model="gpt-neo-2-7b", prompt="Hello there, ", max_length=25, do_sample=True, eos_words=["\n"])
tr = Transformation.prepare(model="punctuation-imputation", texts=["hello my name is tim i just came back from nyc how are you doing"])
pl = Pipeline.prepare(model="en-core-web-lg", texts=["Hi there!"])
# Building and running the asynchronous pipe
asy = AsyncCallPipe([cl, cm, tr, pl])
res = asy.run()
# Fetching the result of the first call in the list
print(res[0].format_result(result_format="pd"))Output:
text label confidence_score
0 I want to kill myself. unsafe 0.9772329101403038
It is possible to add API calls to the asynchronous pipe even after its instantiation though the add function. There's also the flush function to clear the list in the pipe.
from oxapi.asynch import AsyncCallPipe
from oxapi.nlp.encoding import Encoding
# Instantiate an empty asynchornous pipe
asy = AsyncCallPipe()
# Set up API call and add it to the pipe
en = Encoding.prepare(model="mpnet-base-v2", texts=["Hello", "How are you?"])
asy.add(en)
# running the asynchronous pipe
res = asy.run()