The code in this directory defines a client library for use with the gRPC-based Test.ai classifier server.
pip install testai_classifier
This package exposes a ClassifierClient
class:
from testai_classifier import ClassifierClient
You can use it to attempt to match images to a semantic label:
def classify():
client = ClassifierClient(HOST, PORT)
# assume cart_img and menu_img are byte streams as delivered by file.read()
# define a mapping between ids and image data
data = {'cart': cart_img, 'menu': menu_img}
# define which label we are looking to match
label = 'cart'
# attempt to match the images with the label
# confidence is from 0.0 to 1.0 -- any matches with lower than the specified
# confidence are not returned.
# allow_weaker_matches specifies whether to return matches that are above
# the confidence threshold but whose most confident match was a *different*
# label
res = client.classify_images(label, data, confidence=0.0, allow_weaker_matches=True)
# res looks like:
# {'cart': {'label': 'cart', 'confidence': 0.9, 'confidence_for_hint': 0.9},
# 'menu': {'label': 'menu', 'confidence': 0.9, 'confidence_for_hint': 0.2}}
# always close the client connection
client.close()
You can also use it in conjunction with a Selenium Python client driver object, to find elements in a web page based on the label:
def find_elements():
client = ClassifierClient(HOST, PORT)
driver.get("https://test.ai")
els = client.find_elements_matching_label(driver, "twitter")
els[0].click()
assert driver.current_url == "https://twitter.com/testdotai"
client.close()
make install
- install deps (requires Pipenv)make protogen
- generate python client helpers from .proto filemake clean
- reset generated filesmake build
- run setup.py to generate publishable filesmake test
- run test suite (alsomake unit-test
andmake se-test
)make publish
- publish to pypi (alsomake publish-test
)