Python API Service Envirenment. This project offers a web server which offers access to python libraries with communication through JSON.

Take a look at the following piece of python code:

>>> import sklearn.linear_model
>>> reg = sklearn.linear_model.LinearRegression(normalize=True)
>>> reg.fit(X = [[0,0], [1,1], [2,2]], y = [0,1,2])
LinearRegression(copy_X=True, fit_intercept=True, n_jobs=1, normalize=True)
>>> reg.coef_
array([ 0.5,  0.5])
>>> reg.predict(X = [[0.5, 1], [1, 0.5]])
array([ 0.75,  0.75])

The same operations can be realised using a Pase-service. Assume the server is accessible through localhost:5000/.

First create the LinearRegression object reg = sklearn.linear_model.LinearRegression(normalize=True) with the following http request:

POST /sklearn.linear_model.LinearRegression HTTP/1.1
Host: localhost:5000
Content-Type: application/json

{"normalize":true}

The server then returns this JSON-string: {"id": "85F09D2E65", "class": "sklearn.linear_model.base.LinearRegression"}

Now the client can access LinearRegression instance's attributes and methods through this url:

localhost:5000/sklearn.linear_model.base.LinearRegression/85F09D2E65

The equivalent of the reg.fit(X = [[0,0],[1,1],[2,2]], y = [0,1,2]) line would be this http call:

POST /sklearn.linear_model.base.LinearRegression/85F09D2E65/fit HTTP/1.1
Host: localhost:5000
Content-Type: application/json

{"X" : [[0,0],[1,1],[2,2]], "y" : [0,1,2]}

Retrieving the value of reg.coef_ can be done with a GET request:

GET /sklearn.linear_model.base.LinearRegression/85F09D2E65/coef_ HTTP/1.1
Host: localhost:5000

Server response: [0.5, 0.5]

Returning the result of reg.predict(X = [[0.5,1],[1,0.5]]) is done similarly to the fit call from before:

POST /sklearn.linear_model.base.LinearRegression/85F09D2E65/predict HTTP/1.1
Host: localhost:5000
Content-Type: application/json

{"X" : [[0.5, 1], [1, 0.5]]}

Server response: [0.75, 0.75]

To create a Pase-service clone this repository and run bash service.sh

The shell script resolves dependencies and checks if at least Python 3.6 is used. Afterwards it makes the server available on the port specified by the command line parameter. (e.g.: bash service.sh 200)

Make sure libraries that are used by the client are accessible by the server. To create the LinearRegression object from above the scikit-learn framework is needed. One option is to add install commands to the first bracket of service.sh. e.g. adding pip install sklearn makes the scikit-learn framework available.

conf/whitelist.ini contains constructors that are allowed to be created by clients. To whitelist a constructor, add a new line consisting of the fully qualified constructor name followed by a equal sign followed by True.

If debug = True is specified in whitelist.ini, debug mode is enabled. Consequently:

• Flask is started in debug mode.
• No whitelist checks are made. Thus every constructor can be created through http calls.

Say <host> accesses the running pase server.

Creating objects using constructor or class-methods:

method = POST

url = <host>\<fully-qualified-contructor-name>

body = JSON encoded parameters for the constructor. JSON variable names need to be identical to the parameter names in python.

returns = JSON with two values: "id", "class"

Use <host>\<class>\<id> to access the created object.

Retrieving object state:

method = GET

url = <host>\<class>\<id>

returns = JSON encoded state of the object

Retrieve or set the value of a object's attribute (called <attr>):

method = GET (to retrieve value), POST (to retrieve or set value)

url = <host>\<class>\<id>\<attr>

body (if POST) = JSON encoded parameter. Map "value" to the expected value. e.g.: {"value" : 10}

returns = JSON encoded value of the accessed attribute (value after assignment)

Call a function called <func>:

method = POST

url = <host>\<class>\<id>\<func>

body = JSON encoded parameters for the function. JSON variable names need to be identical to the parameter names in python.

returns = JSON encoded return value of the function call

(Note that resulting numpy arrays will be converted to lists (arrays))

Access modifiers are url alterations that modify the server behaviour. Use access modifier names after <class> and before <id> in urls.

To guarantee that the state of the object doesn't change after calling methods or accessing attributes operate like above but use the following urls instead:

<host>\<class>\safe\<id>\<attr>

<host>\<class>\safe\<id>\<func>

Use 'copy' when you want to clone an instance:

GET <host>\<class>\copy\<id>

Use 'copy' while accessing an attribute or calling a method (the same way done above) to copy the (result)value to a new instance object:

<host>\<class>\copy\<id>\<attr>

<host>\<class>\copy\<id>\<func>

The response from a copy access is identical to the response from Creation.