This part of documentation covers answers specifc questions that the user may have.

Where do I find the application config folder?¶

The application folder is located in $HOME/.conML.

Where do I find find the settings files?¶

The settings file is located in $HOME/.conML/settings.ini.

How to read a settings files?¶

Call the conML.load_settings with the destination path of the settings file.

>>> settings = conML.load_settings("settings.ini")

How to create a constructor?¶

Import the desired unsupervised machine learning models.

>>> from sklearn.cluster import KMeans
>>> from sklearn.cluster import AgglomerativeClustering

Instantiate the models and put them in a list with an abbreviation. Both must be stored as tuples.

>>> unsup_models = [("Kme", Kmeans()), ("Agg", AgglormerativeClustering())]

Call the function conML.construction with the the defined list and construction type. Currently only conceptual constructors are supported.

>>> constructor = conML.construction("conceptual", unsup_models)

How to modify the behaviour of a constructor?¶

Every properly instantiated constructor object can be modified via the properties min_category_size and max_categories.

>>> constructor.min_category_size = 0.2
>>> constructor.max_categories = 3

How to create a reconstructor?¶

Import the desired supervised machine learning models.

>>> from sklearn.svm import SVC
>>> from sklearn.neighbors import KNeighborsClassifier

Instantiate the models and put them in a list with an abbreviation as tuple.

>>> sup_models = [("Svc", SVC()), ("Kne", KNeighborsClassifier())]

Call the conML.reconstruction function with the sup_models as argument and the reconstruction type conceptual.

>>> reconstruction = conML.reconstruction("conceptual", sup_models)

How to modify the behaviour of a reconstructor?¶

Every properly instantiated reconstructor object can be modified via the properties min_test_accuracy and reliability_sample.

>>> reconstructor.min_test_accuracy = 0.7
>>> reconstructor.reliability_sample = 0.6

How to create a deconstructor?¶

First create a reconstructor and pass them to the function conML.deconstruction.

>>> from sklearn.svm import SVC
>>> from sklearn.neighbors import KNeighborsClassifier
>>> sup_models = [("Svc", SVC()), ("Kne", KNeighborsClassifier())]
>>> reconstructor = conML.reconstruction("conceptual", sup_models)
>>> deconstructor = conML.deconstruction(reconstructor)

How to modify the behaviour of a deconstructor?¶

Every properly instantiated reconstructor object can be modified via the properties deconst_mode, deconst_strategy, deconst_max_distance_t, deconst_full_tolerance, force_time_expansion, allow_weak_reliability learn_block_minimum and min_reliability.

>>> deconstructor.deconst_mode = "full"
>>> deconstructor.deconst_strategy = "conservative"
>>> deconstructor.deconst_max_distance_t = 0.01
>>> deconstructor.deconst_full_tolerance =  0.1
>>> deconstructor.force_time_expansion = False
>>> deconstructor.allow_weak_reliability = False
>>> deconstructor.learn_block_minumum = 2000
>>> deconstructor.min_reliability = 0.1

How to load a knowledge database?¶

You can load a database by using conML.load_knowledge function. The only argument that is required is the path to the saved database.

>>> database = conML.load_database("knowledge.bin")

How to save a knowledge database?¶

The result of every knowledge search is a KnowledgeDatabase. You should use the save method of the KnowledgeDatabase object.

>>> database.save("knowledge.bin")

How to continue a knowledge search?¶

If you want to continue the knowledge search, you usually start it like any other knowledge search, but with the difference that you pass a knowledge database when you call the context manager.

>>> database = conML.load_database("knowledge.bin")
>>> with conML.knowledge_searcher(*components, database=database) as searcher:
>>>     for block in generate_block():
>>>         database, halde = searcher.search(block)

How to perform the knowledge search in parallel?¶

If you want to perform the knowledge search in parallel, you usually start it like any other knowledge search, but with the difference that you pass a value bigger than 1 to n_procs.

>>> with conML.knowledge_searcher(*components, n_procs=2):
>>>     for block in generate_block():
>>>         database, halde = searcher.search(block)

Where do I find the log files?¶

The log files are located in $HOME/.conML. The log folders have a timestamp as name. The timestamp has the format day, month, year, hour, minute, seconds, for example 31_03_2020_22_53_05.