Python bindings for the Luminoso client API

This package contains Python code for interacting with a Luminoso text processing server through its REST API.

In this code, instead of having to authenticate each request separately, you make a "session" object that keeps track of your login information, and call methods on it that will be properly authenticated.

Installation

This client API is designed to be used with Python 3.

You can download and install it using a Python package manager:

pip install luminoso-api

or

easy_install luminoso-api

Or you can download this repository and install it the usual way:

python setup.py install

If you are installing into the main Python environment on a Mac or Unix system, you will probably need to prefix those commands with sudo and enter your password, as in sudo python setup.py install.

Getting started

You interact with the API using a LuminosoClient object, which sends HTTP requests to URLs starting with a given path, and keeps track of your authentication information.

Before you can connect to an API, you will need to go to the UI on the web and get a long-lived API token. (To get a token, go to the "User settings" option in the upper right dropdown menu, and click the "API tokens" button.) Once you have one, you can use it to connect to the API.

from luminoso_api import V5LuminosoClient
project = V5LuminosoClient.connect('/projects/my_project_id', token='my_token')

# And then, for instance:
docs = project.get('docs', limit=10)

Instead of specifying the token when connecting, you can also use the LuminosoClient to save a token to a file, at which point you can connect without having to specify a token.

from luminoso_api import V5LuminosoClient
V5LuminosoClient.save_token(token='my_token')
project = V5LuminosoClient.connect('/projects/my_project_id')
docs = project.get('docs', limit=10)

Note that the LuminosoClient will ensure that slashes are put in the right places, so that all of the following calls will go to the endpoint https://analytics.luminoso.com/api/v5/projects/my_project_id/docs/:

V5LuminosoClient.connect('/projects/my_project_id').get('docs')
V5LuminosoClient.connect('/projects/my_project_id').get('/docs')
V5LuminosoClient.connect('/projects/my_project_id').get('docs/')
V5LuminosoClient.connect('/projects/my_project_id').get('/docs/')
V5LuminosoClient.connect('/projects/my_project_id/').get('docs')
V5LuminosoClient.connect('/projects/my_project_id/').get('/docs')
V5LuminosoClient.connect('/projects/my_project_id/').get('docs/')
V5LuminosoClient.connect('/projects/my_project_id/').get('/docs/')

HTTP methods

The URLs you can communicate with are documented at https://analytics.luminoso.com/api/v5/. That documentation is the authoritative source for what you can do with the API, and this Python code is just here to help you do it.

A LuminosoClient object has methods such as .get, .post, and .put, which correspond to the corresponding HTTP methods that the API uses. For example, .get is used for retrieving information without changing anything, .post is generally used for creating new things or taking actions, and .put is generally used for updating information.

Examples

Most of the time, you'll want your LuminosoClient to refer to a particular project (also known as a database), but one case where you don't is to get a list of projects in the first place:

from luminoso_api import V5LuminosoClient
client = V5LuminosoClient.connect()
project_info_list = client.get('/projects/')
print(project_info_list)

An example of working with a project, including the use of the convenience method .wait_for_build:

from luminoso_api import V5LuminosoClient
client = V5LuminosoClient.connect()

# Create a new project by POSTing its name and language
project_id = client.post('/projects/', name='testproject', language='en')['project_id']

# use that project from here on
client.change_path('/projects/' + project_id)

docs = [{'title': 'First example', 'text': 'This is an example document.'},
        {'title': 'Second example', 'text': 'Examples are a great source of inspiration.'},
        {'title': 'Third example', 'text': 'Great things come in threes.'}]
client.post('upload', docs=docs)
client.post('build')
client.wait_for_build()

# When the previous call finishes:
response = client.get('concepts')
for concept in response['result']:
    print('%s - %f' % (concept['texts'][0], concept['relevance']))

Vectors

The semantics of terms are represented by "vector" objects, which this API will return as inscrutable base64-encoded strings like this:

'WAB6AJG6kL_6D_6yAHE__R9kSAE8BlgKMo_80y8cCOCCSN-9oAQcABP_TMAFhAmMCUA'

If you want to look inside these vectors and compare them to each other, download our library called pack64, available as pip install pack64. It will turn these into NumPy vectors, so it requires NumPy.

>>> from pack64 import unpack64
>>> unpack64('WAB6AJG6kL_6D_6y')
array([ 0.00046539,  0.00222015, -0.08491898, -0.0014534 , -0.00127411], dtype=float32)

Uploading from the command line

While there is no dedicated command to upload documents, this library does include the command lumi-api, which can be used to access the API in general and to upload documents in particular. Run lumi-api -h for more information.