Artificial Intelligence Management

This is a toolbox to help AI & ML teams to have a better management of their metrics and processes.

Our desire is to enable the company with data related to AI solution, in a easy way to read and use. Some new goals are going to be included later

Confluence Documentation Link

Tangram Link

Table of Contents

• Project Structure
• Features
• Installation/Usage
• Contact

Project Structure

Describe the structure of the project folder, including the organization of modules, directories, and any important files.

ai_management/
├── __init__.py
├── model_evaluation.py
├── config.yaml

Explain the purpose of each module or significant files.

ModelEvaluation

Historize the technical model evaluation results at a Google Big Query table at a Google Cloud Platform project.

Installation

pip install ai-management

Usage

Binary classification

y_true = [1, 0, 0, 1, 1]
y_pred = [1, 0, 0, 0, 1]

y_test_a_lst = y_true
y_pred_a_lst = y_pred

y_test_a_arr = np.array(y_true)
y_pred_a_arr = np.array(y_pred)

Multi class classification

y_true = [0, 1, 2, 1, 2]
y_pred = [[0.9, 0.1, 0.0], [0.3, 0.2, 0.5], [0.2, 0.3, 0.5], [0.1, 0.8, 0.1], [0.1, 0.2, 0.7]]

y_test_b_lst = y_true
y_pred_b_lst = y_pred

y_test_b_arr = np.array(y_true)
y_pred_b_arr = np.array(y_pred)

Multi label classification

y_test = [[0, 1, 2], [3, 4, 5], [6, 7, 8]]
y_pred = [[0, 1, 2], [3, 4, 5], [6, 7, 9]]

y_test_c_lst = y_test
y_pred_c_lst = y_pred

y_test_c_arr = np.array(y_true)
y_pred_c_arr = np.array(y_pred)

Regression

y_true = [2.5, 3.0, 4.0, 5.5, 6.0]
y_pred = [2.0, 3.5, 3.8, 5.0, 6.5]

y_test_d_lst = y_true
y_pred_d_lst = y_pred

y_test_d_arr = np.array(y_true)
y_pred_d_arr = np.array(y_pred)

Assossiation Rules

import pandas as pd
import numpy as np

# Create a dataframe with random values
df_assossiation = pd.DataFrame({
    'ID_PRNCPAL': np.random.randint(1, 50000, size=103846),
    'CONFIDENCE': np.random.uniform(0.01, 0.03, size=103846)
})

df_assossiation.sort_values('ID_PRNCPAL')

Solution Evaluation

import ai_management as aim 
client_bq = bigquery.Client(project='project')
me = aim.ModelEvaluation(
    client_bq=client_bq,
    destination='project.dataset.table'
)

# Historizing standard metrics
me.historize_model_evaluation(
    soltn_nm = 'Solution X', 
    lst_mdls = [
        {
            'mdl_nm' : 'Model A',
            'algrthm_typ' : 'binary_classification',
            'data' : [y_test_a_lst, y_pred_a_lst]}, 
        {
            'mdl_nm' : 'Model B',
            'algrthm_typ' : 'multi_class_classification',
            'data' : [y_test_b_lst, y_pred_b_lst]},
        {
            'mdl_nm' : 'Model C',
            'algrthm_typ' : 'multi_label_classification',
            'data' : [y_test_c_lst, y_pred_c_lst]},
        {
            'mdl_nm' : 'Model D',
            'algrthm_typ' : 'assossiation',
            'data' : ['confidence', df_assossiation]},
    ]
)

# Historizing custom metrics
me.historize_custom_metric(
    soltn_nm = "Solution Y",
    lst_mdls = [
        {
            'mdl_nm': 'Model E',
            'algrthm_typ': 'regression',
            'data': [
                ["Lin's Concordance Correlation Coefficient", 0.85, None],
                ["Huber's error", 123, {"delta": 0.75}],
            ]
        },
    ]
)

Contact

• Leroy Merlin Brazil AI scientists and developers: chapter_inteligencia_artificia@leroymerlin.com.br