NAV
  • Python SDK
  • Python SDK

    Introduction

    Welcome to the developer documentation for the Kameleoon Python SDK! Our SDK gives you the possibility of running experiments and activating feature flags on your back-end Python server. Integrating our SDK into your web-application is easy, and its footprint (in terms of memory and network usage) is low.

    You can refer to the SDK reference to check out all possible features of the SDK. Also make sure you check out our Getting started tutorial which we have prepared to walk you through the installation and implementation.

    Latest version of the Python SDK: 1.0.4.

    Getting started

    This guide is designed to help you integrate our SDK in a few minutes and start running experiments in your Python applications. This tutorial will explain the setup of a simple A/B test to change the number of recommended products based on different variations.

    Creating an experiment

    First, you must create an experiment in the Kameleoon back-office so that our platform is aware of the new A/B test you're planning to implement on your side. Make sure that server-side type is chosen as shown below:

    Server-side experiment

    Upon successful creation of the experiment, you will need to get its ID to use in the SDK as an argument to the trigger_experiment() method.

    Installing the SDK

    Installing the Python client

    pip install kameleoon-client-python
    

    Installing the SDK can be directly achieved through an Python pip package. Our package is hosted on the official pip repository, so you just have to run the following command:

    pip install kameleoon-client-python
    

    Additional configuration

    You should provide credentials for the Python SDK via a configuration file, which can also be used to customize the SDK behavior. A sample configuration file can be obtained here. We suggest to install this file to the default path of /etc/kameleoon/client-python.yaml, but you can also put it in another location and passing the path as an argument to the KameleoonClient() constructor method. With the current version of the Python SDK, those are the available keys:

    Initializing the Kameleoon client

    from kameleoon import KameleoonClient
    
    SITE_CODE = 'a8st4f59bj'
    
    kameleoon_client = KameleoonClient(SITE_CODE)
    
    kameleoon_client = KameleoonClient(SITE_CODE, blocking=False, configuration_path='/etc/kameleoon/client-python.yaml')
    
    # Second version of the Kameleoon::Client, using an asynchronous trigger_experiment() method
    kameleoon_client = KameleoonClient(SITE_CODE, blocking=True)
    
    kameleoon_client = KameleoonClient(SITE_CODE, blocking=True, logger=MyLogger)
    

    After installing the SDK into your application, configuring the correct credentials (in /etc/kameleoon/client-python.yaml) and setting up a server-side experiment on Kameleoon's back-office, the next step is to create the Kameleoon client in your application code.

    The code on the right gives a clear example. A Kameleoon::Client is a singleton object that acts as a bridge between your application and the Kameleoon platform. It includes all the methods and properties you will need to run an experiment.

    Using the Kameleoon Python SDK in a Django environment

    If you use Django, we recommend you to initialize the Kameleoon client at server start-up, in the apps.py file of your Django application.

    When you use python manage.py runserver Django start two processes, one for the actual development server and other to reload your application when the code change.

    You can also start the server without the reload option, and you will see only one process running will only be executed once :

    python manage.py runserver --noreload

    You can also just check the RUN_MAIN env var in the ready() method itself.

    def ready(self): if os.environ.get('RUN_MAIN', None) == 'true': configuration_path = os.path.join(ROOT_DIR, 'path_to_config', 'config.yml') self.kameleoon_client = KameleoonClient(SITE_CODE, configuration_path=configuration_path)

    This only applies to local development when you use python manage.py runserver. In a production environment, the code in the ready() function will be executed only one time when the application is initialized.

    from django.apps import apps
    
      my_application = apps.get_app_config('your_app')
      client = my_application.kameleoon_client
    

    You can then access the Kameleoon client in your application.

    Triggering an experiment

    
    # If you don't use Django, but you're working in a web environment and want to use the visitor code via a cookie, you must provide custom implementations of the read_cookies() and write_cookies() methods.
    
    visitor_code = kameleoon_client.obtain_visitor_code(request.COOKIES, 'example.com')
    
    variation_id = 0
    
    try:
        variation_id = kameleoon_client.trigger_experiment(visitor_code, 135471)
    except NotActivated as ex:
        # The user triggered the experiment, but did not activate it.
        # Usually, this happens because the user has been associated
        # with excluded traffic
        variation_id = 0
        client.logger.error(ex)
    except NotTargeted as ex:
        # The user did not trigger the experiment, as the associated
        # targeting segment conditions were not fulfilled.
        # He should see the reference variation
        variation_id = 0
        client.logger.error(ex)
    except ExperimentConfigurationNotFound as ex:
        # The user will not be counted into the experiment,
        # but should see the reference variation
        variation_id = 0
        client.logger.error(ex)
    
    recommended_products_number = 5
    # This is the default / reference number of products to display
    
    if variation_id == 148382:
        # We are changing number of recommended products for this variation to 10
        recommended_products_number = 10
    elif variation_id == 187791:
        # We are changing number of recommended products for this variation to 8
        recommended_products_number = 8
    
    # Here you should have code to generate the HTML page back to the client, where recommendedProductsNumber will be used
    response = JsonResponse({...})
    

    Running an A/B experiment on your Python application means bucketing your visitors into several groups (one per variation). The SDK takes care of this bucketing (and the associated reporting) automatically.

    Triggering an experiment by calling the trigger_experiment() method will register a random variation for a given visitor_code. If this visitor_code is already associated with a variation (most likely a returning visitor that has already been exposed to the experiment previously), then it will return the previous variation assigned for the given experiment.

    Implementing variation code

    if variation_id == 0
        # This is the default / reference number of products to display
        recommended_products_number = 5
    elsif variation_id == 148382
        # We are changing number of recommended products for this variation to 10
        recommended_products_number = 10
    elsif variation_id == 187791
        # We are changing number of recommended products for this variation to 8
        recommended_products_number = 8
    end
     # Here you should have code to generate the response back to the client, where recommended_products_number will be used.
    

    To execute different code paths depending on the variation assigned to the visitor, you will need the list of all the experiment's variation IDs. You can find these variation IDs (as well as the experiment ID) by opening the experiment in the back-office interface. By convention, the reference (original variation) always has an ID equal to 0.

    Once you have the IDs of the different variations, you can implement a different action for each variation, and one of the code paths will be executed, based on the associated variation_id for the current visitor. Generally, this can be done using a simple if / else or switch mechanism. In our example, we just change the number of recommended products with two different variations.

    Get variationID

    Tracking conversion

    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com")
    goal_id = 83023
    
    kameleoon_client.track_conversion(visitor_code, goal_id);
    

    After you are done triggering an experiment, the next step is usually to start tracking conversions. This is done to measure performance characteristics according to the goals that make sense for your business.

    For this purpose, use the track_conversion() method of the SDK as shown in the example. You need to pass the visitor_code and goal_id parameters so we can correctly track conversions for this particular visitor.

    Get goalID

    Obtaining results

    Once your implementation is in place on the server side (experiment triggering, variations handling, and conversion tracking), it is time to launch the experiment on the Kameleoon platform. You do this in the same way as for a front-end test. Basic operations such as starting, pausing and stopping the experiment work exactly the same way.

    After the experiment is launched, the first results will be available on our standard results page in the back-office after a duration of 30 minutes. This is because (as is the case with front-end testing) visits are considered over after 30 minutes of inactivity. Inactivity in this context means the absence of calls sent to the Kameleoon back-end servers (such calls are made via trigger_experiment(), track_conversion() or flush() methods).

    Technical considerations

    Kameleoon made an important architectural design decision with its SDK technology, namely that every SDK must comply with a zero latency policy. In practice, this means that any blocking remote server call is banned, as even the fastest remote call would add a 20ms latency to your application. And if for any reason our servers are slower to reply than usual (unfortunately, this can happen), this delay can quickly increase to hundreds of milliseconds, seconds... or even completely block the load of the web page for the end user. We believe that web performance is of paramount importance in today's world and we don't think adding server-side A/B testing or feature flagging capabilities should come at the cost of an increased web page rendering time. For this reason, we guarantee that the use of our SDKs has absolutely no impact on the performance of the host platform.

    However, having a zero latency policy does impose some constraints. The main one is that user data about your visitor should be kept locally, and not fetched from a remote server. For instance, if you set a custom data for a given visitor, we must store this somehow in your server / infrastructure, not on our (remote) side.

    In the case of the Python SDK, this is implemented via a hash of visitor data (where keys are the visitor_codes) directly on RAM. So if you use Kameleoon::CustomData.new() and then kameleoon_client.add_data() methods, the information will be stored in the application's RAM (the one hosting the SDK, usually an application server). The map is regularly cleaned (old visitors data is erased) so it should usually not grow too big in size, unless you have a very big traffic and use lots of custom data.

    Reference

    This is a full reference documentation of the Python SDK.

    If this is your first time working with the Python SDK, we strongly recommend you go over our Getting started tutorial to integrate the SDK and start experimenting in a few minutes.

    kameleoon.KameleoonClient

    init

    from kameleoon import KameleoonClient
    
    SITE_CODE = 'a8st4f59bj'
    
    kameleoon_client = KameleoonClient(SITE_CODE)
    
    kameleoon_client = KameleoonClient(SITE_CODE, blocking=False, configuration_path='/etc/kameleoon/client-python.yaml')
    

    The starting point for using the SDK is the initialization step. All interactions with the SDK are done through an object named Kameleoon::Client, therefore you need to create this object.

    Arguments
    NameTypeDescription
    site_codestrCode of the website you want to run experiments on. This unique code id can be found in our platform's back-office. This field is mandatory.
    blockingboolThis parameter defines if the trigger_experiment() method has a non-blocking or blocking behavior. Value True will set it to be blocking. This field is optional and set to False by default.
    configuration_file_pathstrPath to the SDK configuration file. This field is optional and set to /etc/kameleoon/client-python.yaml by default.
    client_idStringThis parameter is used for OAUth 2.0 authentication to our service. This field is optional, as it can be provided via the configuration file. However, it must either be supplied by the configuration file or by this method, else a Kameleoon::Exception::CredentialsNotFound exception will be thrown.
    client_secretStringThis parameter is used for OAUth 2.0 authentication to our service. This field is optional, as it can be provided via the configuration file. However, it must either be supplied by the configuration file or by this method, else a Kameleoon::Exception::CredentialsNotFound exception will be thrown.

    Kameleoon::Client

    obtain_visitor_code

    require "securerandom"
    
    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com")
    
    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com", visitor_code)
    
    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com", SecureRandom.uuid)
    

    The obtain_visitor_code() helper method should be called to obtain the Kameleoon visitor_code for the current visitor. This is especially important when using Kameleoon in a mixed front-end and back-end environment, where user identification consistency must be guaranteed. The implementation logic is described here:

    1. First we check if a kameleoonVisitorCode cookie or query parameter associated with the current HTTP request can be found. If so, we will use this as the visitor identifier.

    2. If no cookie / parameter is found in the current request, we either randomly generate a new identifier, or use the default_visitor_code argument as identifier if it is passed. This allows our customers to use their own identifiers as visitor codes, should they wish to. This can have the added benefit of matching Kameleoon visitors with their own users without any additional look-ups in a matching table.

    3. In any case, the server-side (via HTTP header) kameleoonVisitorCode cookie is set with the value. Then this identifier value is finally returned by the method.

    For more information, refer to this article.

    Arguments
    NameTypeDescription
    cookiesHashCookies on the current HTTP request should be passed, as a Hash object ({:cookie_name => cookie_value}). This field is mandatory.
    top_level_domainStringYour current top level domain for the concerned site (this information is needed to set the corresponding cookie accordingly, on the top level domain). This field is mandatory.
    default_visitor_codeStringThis parameter will be used as the visitor_code if no existing kameleoonVisitorCode cookie is found on the request. This field is optional, and by default a random visitor_code will be generated.
    Return value
    TypeDescription
    StringA visitor_code that will be associated with this particular user and should be used with most of the methods of the SDK.

    trigger_experiment

    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com")
    experiment_id = 75253
    
    begin
        variation_id = kameleoon_client.trigger_experiment(visitor_code, experiment_id)
    rescue Kameleoon::Exception::NotTargeted
        # The user did not trigger the experiment, as the associated targeting segment conditions were not fulfilled. He should see the reference variation
        variation_id = 0
    rescue Kameleoon::Exception::NotActivated
        # The user triggered the experiment, but did not activate it. Usually, this happens because the user has been associated with excluded traffic
        variation_id = 0 
    rescue Kameleoon::Exception::ExperimentConfigurationNotFound
        # The user will not be counted into the experiment, but should see the reference variation
        variation_id = 0 
    end
    

    To trigger an experiment, call the trigger_experiment() method of our SDK.

    This method takes visitor_code and experiment_id as mandatory arguments to register a variation for a given user.

    If such a user has never been associated with any variation, the SDK returns a randomly selected variation. If a user with a given visitor_code is already registered with a variation, it will detect the previously registered variation and return the variation_id.

    You have to make sure that proper error handling is set up in your code as shown in the example to the right to catch potential exceptions.

    Arguments
    NameTypeDescription
    visitor_codeStringUnique identifier of the user. This field is mandatory.
    experiment_idIntegerID of the experiment you want to expose to a user. This field is mandatory.
    timeoutIntegerTimeout (in milliseconds). This parameter is only used in the blocking version of this method, and specifies the maximum amount of time the method can block to wait for a result. This field is optional. If not provided, it will use the default value of 2000 milliseconds.
    Return value
    NameTypeDescription
    variation_idIntegerID of the variation that is registered for a given visitor_code. By convention, the reference (original variation) always has an ID equal to 0.
    Exceptions Thrown
    Error MessageDescription
    Kameleoon::Exception::NotTargetedException indicating that the current visitor / user did not trigger the required targeting conditions for this experiment. The targeting conditions are defined via Kameleoon's segment builder.
    Kameleoon::Exception::NotActivatedException indicating that the current visitor / user triggered the experiment (met the targeting conditions), but did not activate it. The most common reason for that is that part of the traffic has been excluded from the experiment and should not be tracked.
    Kameleoon::Exception::ExperimentConfigurationNotFoundException indicating that the requested experiment ID has not been found in the internal configuration of the SDK. This is usually normal and means that the experiment has not yet been started on Kameleoon's side (but code triggering / implementing variations is already deployed on the web-application's side).

    activate_feature

    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com")
    
    feature_key = "new_checkout"
    has_new_checkout = false
    
    begin
        has_new_checkout = kameleoon_client.activate_feature(visitor_code, feature_key)
    rescue Kameleoon::Exception::NotTargeted
        # The user did not trigger the feature, as the associated targeting segment conditions were not fulfilled. The feature should be considered inactive
        has_new_checkout = false
    rescue Kameleoon::Exception::FeatureConfigurationNotFound
        # The user will not be counted into the experiment, but should see the reference variation
        has_new_checkout = false
    end
    
    if has_new_checkout
        # Implement new checkout code here
    end
    
    

    To activate a feature toggle, call the activate_feature() method of our SDK.

    This method takes a visitor_code and feature_key (or feature_id) as mandatory arguments to check if the specified feature will be active for a given user.

    If such a user has never been associated with this feature flag, the SDK returns a boolean value randomly (true if the user should have this feature or false if not). If a user with a given visitor_code is already registered with this feature flag, it will detect the previous featureFlag value.

    You have to make sure that proper error handling is set up in your code as shown in the example to the right to catch potential exceptions.

    Arguments
    NameTypeDescription
    visitor_codeStringUnique identifier of the user. This field is mandatory.
    feature_id or feature_keyInteger or StringID or Key of the feature you want to expose to a user. This field is mandatory.
    timeoutIntegerTimeout (in milliseconds). This parameter is only used in the blocking version of this method, and specifies the maximum amount of time the method can block to wait for a result. This field is optional, if not provided, it will use the default value of 2000 milliseconds.
    Return value
    TypeDescription
    BooleanValue of the feature that is registered for a given visitor_code.
    Exceptions Thrown
    TypeDescription
    Kameleoon::Exception::NotTargetedException indicating that the current visitor / user did not trigger the required targeting conditions for this feature. The targeting conditions are defined via Kameleoon's segment builder.
    Kameleoon::Exception::FeatureConfigurationNotFoundException indicating that the requested feature ID has not been found in the internal configuration of the SDK. This is usually normal and means that the feature flag has not yet been activated on Kameleoon's side (but code implementing the feature is already deployed on the web-application's side).

    obtain_variation_associated_data

    require "json"
    
    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com")
    experiment_id = 75253
    
    begin
        variation_id = kameleoon_client.trigger_experiment(visitor_code, experiment_id)
        json_object = kameleoon_client.obtain_variation_associated_data(variation_id) # Return a json encoded string
        first_name = json_object["firstName"]
    rescue Kameleoon::Exception::VariationConfigurationNotFound
        # The variation is not yet activated on Kameleoon's side, ie the associated experiment is not online
    end
    
    

    To retrieve JSON data associated with a variation, call the obtain_variation_associated_data() method of our SDK. The JSON data usually represents some metadata of the variation, and can be configured on our web application interface or via our Automation API.

    This method takes the variation_id as a parameter and will return the data as a Hash instance. It will throw an exception (Kameleoon::Exception::VariationConfigurationNotFound) if the variation ID is wrong or corresponds to an experiment that is not yet online.

    Arguments
    NameTypeDescription
    variation_idIntegerID of the variation you want to obtain associated data for. This field is mandatory.
    Return value
    TypeDescription
    HashData associated with this variationID.
    Exceptions Thrown
    TypeDescription
    Kameleoon::Exception::VariationConfigurationNotFoundException indicating that the requested variation ID has not been found in the internal configuration of the SDK. This is usually normal and means that the variation's corresponding experiment has not yet been activated on Kameleoon's side.

    obtain_feature_variable

    feature_key = "myFeature"
    variable_key = "myVariable"
    
    begin
        data = kameleoon_client.obtain_feature_variable(feature_key, variable_key)
    rescue Kameleoon::Exception::FeatureConfigurationNotFound
        # The feature is not yet activated on Kameleoon's side
    rescue Kameleoon::Exception::FeatureVariableNotFound
        # Request variable not defined on Kameleoon's side
    end
    
    

    To retrieve a feature variable, call the obtain_feature_variable() method of our SDK. A feature variable can be changed easily via our web application.

    This method takes two input parameters: feature_key and variable_key. It will return the data with the expected type, as defined on the web interface. It will throw an exception (Kameleoon::Exception::FeatureConfigurationNotFound) if the requested feature has not been found in the internal configuration of the SDK.

    Arguments
    NameTypeDescription
    feature_id or feature_keyInteger or StringID or Key of the feature you want to obtain to a user. This field is mandatory.
    variable_keyStringKey of the variable. This field is mandatory.
    Return value
    TypeDescription
    Integer of String or Boolean or HashData associated with this variable for this feature flag. This can be a Integer, String, Boolean or Hash (depending on the type defined on the web interface).
    Exceptions Thrown
    TypeDescription
    Kameleoon::Exception::FeatureConfigurationNotFoundException indicating that the requested feature ID has not been found in the internal configuration of the SDK. This is usually normal and means that the feature flag has not yet been activated on Kameleoon's side.
    Kameleoon::Exception::FeatureVariableNotFoundException indicating that the requested variable has not been found. Check that the variable's ID (or key) matches the one in your code.

    track_conversion

    require "kameleoon"
    require "kameleoon/data"
    
    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com")
    goal_id = 83023
    
    kameleoon_client.add_data(visitor_code, Kameleoon::Browser.new(Kameleoon::BrowserType::CHROME))
    kameleoon_client.add_data(
        visitor_code,
        Kameleoon::PageView.new("https://url.com", "title", 3),
        Kameleoon::Interest.new(2)
    )
    kameleoon_client.add_data(visitor_code, Kameleoon::Conversion.new(32, 10, false))
    kameleoon_client.track_conversion(visitor_code, goal_id)
    

    To track conversion, use the track_conversion() method. This method requires visitor_code and goal_id to track conversion on this particular goal. In addition, this method also accepts revenue as a third optional argument to track revenue. The visitor_code is usually identical to the one that was used when triggering the experiment.

    The track_conversion() method doesn't return any value. This method is non-blocking as the server call is made asynchronously.

    Arguments
    NameTypeDescription
    visitor_codeStringUnique identifier of the user. This field is mandatory.
    goal_idIntegerID of the goal. This field is mandatory.
    revenueFloatRevenue of the conversion. This field is optional.

    addData

    require "kameleoon"
    require "kameleoon/data"
    
    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com")
    
    kameleoon_client.add_data(visitor_code, Kameleoon::Browser.new(Kameleoon::BrowserType::CHROME))
    kameleoon_client.add_data(
        visitor_code,
        Kameleoon::PageView.new("https://url.com", "title", 3),
        Kameleoon::Interest.new(0)
    )
    kameleoon_client.add_data(visitor_code, Kameleoon::Conversion.new(32, 10, false))
    

    To associate various data with the current user, we can use the add_data() method. This method requires the visitor_code as a first parameter, and then accepts several additional parameters. These additional parameters represent the various Data Types allowed in Kameleoon.

    Note that the add_data() method doesn't return any value and doesn't interact with the Kameleoon back-end servers by itself. Instead, all declared data is saved for further sending via the flush() method described in the next paragraph. This reduces the number of server calls made, as data is usually grouped into a single server call triggered by the execution of flush().

    Arguments
    NameTypeDescription
    visitor_codeStringUnique identifier of the user. This field is mandatory.
    data_typesKameleoonDataCustom data types which may be passed separated by a comma.

    flush

    require "kameleoon"
    require "kameleoon/data"
    
    visitor_code = kameleoon_client.obtain_visitor_code(cookies, "example.com")
    
    kameleoon_client.add_data(visitor_code, Kameleoon::Browser.new(Kameleoon::BrowserType::CHROME))
    kameleoon_client.add_data(
        visitor_code,
        Kameleoon::PageView.new("https://url.com", "title", 3),
        Kameleoon::Interest.new(0)
    )
    kameleoon_client.add_data(visitor_code, Kameleoon::Conversion.new(32, 10, false))
    
    kameleoon_client.flush(visitor_code)
    
    

    Data associated with the current user via add_data() method is not immediately sent to the server. It is stored and accumulated until it is sent automatically by the trigger_experiment() or track_conversion() methods, or manually by the flush() method. This allows the developer to control exactly when the data is flushed to our servers. For instance, if you call the add_data() method a dozen times, it would be a waste of ressources to send data to the server after each add_data() invocation. Just call flush() once at the end.

    The flush() method doesn't return any value. This method is non-blocking as the server call is made asynchronously.

    Arguments
    NameTypeDescription
    visitor_codeStringUnique identifier of the user. This field is mandatory.

    Kameleoon::Data

    Browser

    kameleoon_client.add_data(visitor_code, Kameleoon::Browser.new(Kameleoon::BrowserType::CHROME))
    
    NameTypeDescription
    browserKameleoon::BrowserTypeList of browsers: CHROME, INTERNET_EXPLORER, FIREFOX, SAFARI, OPERA, OTHER. This field is mandatory.

    PageView

    kameleoon_client.add_data(visitor_code, Kameleoon::PageView.new("https://url.com", "title", 3))
    
    NameTypeDescription
    urlStringURL of the page viewed. This field is mandatory.
    titleStringTitle of the page viewed. This field is mandatory.
    referrerIntegerIndex of referrer. This field is optional.

    Conversion

    kameleoon_client.add_data(visitor_code, Kameleoon::Conversion.new(32, 10, false))
    
    NameTypeDescription
    goal_idIntegerID of the goal. This field is mandatory.
    revenueFloatConversion revenue. This field is optional.
    negativeBooleanDefines if the revenue is positive or negative. This field is optional.

    CustomData

    kameleoon_client.add_data(visitor_code, Kameleoon::CustomData.new(1, "some custom value"))
    
    NameTypeDescription
    indexIntegerIndex / ID of the custom data to be stored. This field is mandatory.
    valueStringValue of the custom data to be stored. This field is mandatory.

    Interest

    kameleoon_client.add_data(visitor_code, Kameleoon::Interest.new(0))
    
    NameTypeDescription
    indexIntegerIndex of interest. This field is mandatory.