bixai 0.1.9

Package that helps you make some machine learning models easier to understand

bixai

The bixai is a package made for NLO to make an attempt at understanding drivers behind models

It contains multiple modules to analyse different problems:

1. Decomposition for logistic regressions over time
2. Multinomial logistic regression with impact of variables
3. Decision tree / beslisboom
4. Time-series forecasting with regressions and random forest

Installation

Use the package manager pip to install bixai.

pip install bixai

Usage Decomposition for logistic regressions over time

# import modules
from bixai.generate_example_data import GenerateData
from bixai.creating_dataset import CreatingDataSet
from bixai.logistic_regression_decomp import LogisticDecomposition
from random import randint
from sklearn.linear_model import LogisticRegression
import plotly.io as pio
pio.renderers.default = 'browser'

# Generate data
example_data = GenerateData(10000)
df_example = example_data.generate_dataset()

# Using CreatingDataSet to clean the data
getting_data = CreatingDataSet(df_example, {})

# The sample size you want from the data and the variables (X and y) to be used
subset_size = 5000 
X_vars = ['percentage_gelezen_mails', 'geslacht', 'leeftijd', 'maanden_lid', 'kanaal_instroom', 'actie_instroom',
          'contact_vorm']
y = 'churn'

# Create the test/train
X_train, X_test, y_train, y_test = getting_data.get_train_test(y, X_vars, divided_by_max=False, scale_data=True,
                                                               add_random_int=False, add_random_cont=False, set_seed=2,
                                                               size=subset_size, test_size=0.25, random_state=12,
                                                               with_mean=True, with_std=True)

# Logistic Regression Model
model_churn = LogisticRegression().fit(X_train, y_train)

# Add a random variable to split on for this example (this should come from own data)
split_var = [randint(2018, 2022) for p in range(0, len(X_train))] 
vars_to_show = []
model_decomp = LogisticDecomposition(model_churn)
decomposition_results = model_decomp.decomposition_logistic(X_train, split_var=split_var, plot=True, y=[],
                                                             X_vars_to_show=vars_to_show,
                                                             scaling_to_mean_odds_model=True,
                                                             scaling_to_actual_odds=False)

Usage Multinomial Logistic Regression

from bixai.generate_example_data import GenerateData
from bixai.creating_dataset import CreatingDataSet
from bixai.ml_models import Models, EvaluateModels
import plotly.io as pio
pio.renderers.default = 'browser'

# Generate data
example_data = GenerateData(10000)
df_example = example_data.generate_dataset()

# Select what and how much data we wanna use
getting_data = CreatingDataSet(df_example, {})
subset_size = 10000
# Selection of variables we wanna use in the model:
X_vars = ['percentage_gelezen_mails', 'geslacht', 'leeftijd', 'maanden_lid', 'kanaal_instroom', 'actie_instroom',
          'contact_vorm']
# The variable we wanna explain:
y = 'tweede_merk_keuze_stl_spelers'
# Get train/test datasets
X_train, X_test, y_train, y_test = getting_data.get_train_test(y, X_vars, divided_by_max=False, scale_data=True,
                                                               add_random_int=False, add_random_cont=False, set_seed=2,
                                                               size=subset_size, test_size=0.25, random_state=12,
                                                               with_mean=True, with_std=True)
# Multinomial Logistic Regression Model
model = Models(X_train, y_train)
model_mvl = model.multivariate_logistic_regression()

# Get the accuracy of the model
evalueren_model = EvaluateModels(X_train, X_test, y_train, y_test)
evalueren_model.accuracy_models([model_mvl])

# Visualize the probabilities of the model
evalueren_model.visualize_probabilities_mvlogit(model_mvl, bin_size=0.01).show()

# Visualize the impact of each variable on a brand, merken_x_as=True gives the brands on the x_as, False the y_as
evalueren_model.visualize_impact_variables(model_mvl, merken_x_as=False).show()

Usage Decision tree / beslisboom

from functions.generate_example_data import GenerateData
from functions.make_html_simon import CreateHTML
import time

# Generate some sample data
example_data = GenerateData(10000)
df_example = example_data.generate_dataset()

# Input for the tree
variables_ = ['leeftijd', 'geslacht', 'kanaal_instroom']  # The variables we want in the tree
reorder = True  # False als de volgorde moet zijn zoals in variables_, anders op meest impactvolle split op gini/mean
y = 'churn'  # de veriabelen waarvan de mean telkens wordt berekend
split_method = 'gini'  # de methode waarop gesplits kan worden (gini of mean)
min_records = 500  # min N waarna nog een split gemaakt wordt
max_integer = 5  # maximaal aantal splits bij een integer variabelen
max_nr_splits = 2  # behalve voor categorische variabelen
min_split_values = 1000  # minimale N voor een split
nr_splits = {'leeftijd': 4}  # aantal splits per variabelen (kan overschreven worden door splits)
splits = {'leeftijd': [20,25,40,60]}  # op welke waarde een split
color_reverse = True  # omkering kleuren. bij True, rood laag, blauw hoog
name_all = 'Alle spelers'  # De naam die bij het eerste bolletje staat

# Create the actual HTML file
create_html = CreateHTML(df_example, variables_, y, split_method=split_method, min_records=min_records,
                         max_integer=max_integer, max_nr_splits=max_nr_splits, min_split_values=min_split_values,
                         nr_splits=nr_splits, splits=splits, color_reverse=color_reverse, name_all=name_all,
                         reorder=reorder)
start = time.process_time()

# Input for the created HTML filel
output_file = 'beslisboom_voorbeeld.html'  # name you wanna give it (has to end with .html)
title = 'Super Insights'  # The title on top of the html
explanation= 'One Planet, Plant it:</br> <span class="emoji">&#128514;</span>' # Text for in the explanation box
made_by = 'Een toppertje van BI'  # Made by in the left corner of the file
create_html.build_HTML(output_file=output_file, title=title, explanation=explanation, made_by=made_by)
print(time.process_time() - start)

License

Copyright (c) 2023 Rumiko Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.