python-mltk 0.1.4

Machine Learning Toolkit (MLToolkit/mltk) for Python

MLToolkit

Current release: PyMLToolkit [v0.1.5(prev)]

MLToolkit (mltk) is a Python package providing a set of user-friendly functions to help building machine learning models in data science research, teaching or production focused projects.

Introduction

MLToolkit supports all stages of the machine learning application development process.

Installation

pip install python-mltk

If the installation failed with dependancy issues, execute the above command with --no-dependencies

pip install python-mltk --no-dependencies

Functions

• Data Extraction (SQL, Flatfiles, etc.)
• Exploratory Data Analysis (statistical summary, univariate analysis, etc.)
• Feature Engineering
• Model Building (Currently supported for binary classification only)
• Hyper Parameter Tuning [in development for v0.2]
• Model Performance Analysis and Comparison Between Models
• Auto ML (automated machine learning) [in development for v0.2]
• Model Deploymet and Serving [will be imporved for v0.2]

Supported Machine Learning Algorithms/Packages

• RandomForestClassifier: scikit-learn
• LogisticRegression: statsmodels
• Deep Feed Forward Neural Network (DFF): tensorflow
• Convlutional Neural Network (CNN): tensorflow
• Gradient Boost : catboost
• ... More models will be added in the future releases ...

Usage

import mltk

Warning: Python Variable, Function or Class names

The Python interpreter has a number of built-in functions. It is possible to overwrite thier definitions when coding without any rasing a warning from the Python interpriter. (https://docs.python.org/3/library/functions.html) Therfore, AVOID THESE NAMES as your variable, function or class names.

abs	all	any	ascii	bin	bool	bytearray	bytes
callable	chr	classmethod	compile	complex	delattr	dict	dir
divmod	enumerate	eval	exec	filter	float	format	frozenset
getattr	globals	hasattr	hash	help	hex	id	input
int	isinstance	issubclass	iter	len	list	locals	map
max	memoryview	min	next	object	oct	open	ord
pow	print	property	range	repr	reversed	round	set
setattr	slice	sorted	staticmethod	str	sum	super	tuple
type	vars	zip	__import__

If you accedently overwrite any of the built-in function (e.g. list), execute the following to bring built-in defition.

del(list)

Similarly, avoid using special charcters and spaces in the column names of the DataFrames. Execute the following to remove special characters from the column names.

Data = mltk.clean_column_names(Data, replace='')

MLToolkit Example

Data Loading and exploration

import numpy as np
import pandas as pd
import mltk as mltk

Data = pd.read_csv(r'incomedata.csv')
Data = mltk.clean_column_names(Data, replace='')
Data = mltk.add_identity_column(Data, id_label='ID', start=1, increment=1)
DataStats = mltk.data_description(Data)

Data Pre-processing and Feature Engineering

# Analyze Response Target
print(mltk.variable_frequency(DataFrame=Data, variable='income'))

# Set Target Variables
targetVariable = 'HighIncome'
targetCondition = "income=='>50K'" #For Binary Classification

Data=mltk.set_binary_target(Data, target_condition=targetCondition, target_variable=targetVariable)
print(mltk.variable_frequency(DataFrame=Data, variable=targetVariable))

        Counts  CountsFraction%
income                         
<=50K    24720         75.91904
>50K      7841         24.08096
TOTAL    32561        100.00000

# Flag Records to Exclude
excludeCondition="age < 18"
action = 'flag' # 'drop' #
excludeLabel = 'EXCLUDE'
Data=mltk.exclude_records(Data, exclude_ondition=excludeCondition, action=action, exclude_label=excludeLabel) # )#

# Get list of uniques values in categorical variables
categoryVariables = set({'sex', 'nativecountry', 'race', 'occupation', 'workclass', 'maritalstatus', 'relationship'})
print(mltk.category_lists(Data, list(categoryVariables)))

# Merge unique categorical values
groups = [{'variable':'maritalstatus', 'group_name':'Married', 'values':["Married-civ-spouse", "Married-spouse-absent", "Married-AF-spouse"]}]
Data = mltk.merge_categories(Data, groups)

# Show Frequency distribution of categorical variable
sourceVariable='maritalstatus'
table = mltk.variable_frequency(Data, variable=sourceVariable, show_plot=False)
table.style.background_gradient(cmap='Greens').set_precision(3)

# Response Rate For Categorical Variables
mltk.variable_responses(Data, variables=categoryVariables, target_variable=targetVariable, show_output=False, show_plot=True)

# Create Categorical Variables from continious variables
sourceVariable='age'
table = mltk.histogram(Data, sourceVariable, n_bins=10, orientation='vertical', show_plot=True)
print(table)

# Divide to categories
labels = ['0', '20', '30', '40', '50', '60', 'INF']
Data, groupVariable = mltk.numeric_to_category(DataFrame=Data, variable=sourceVariable, str_labels=labels, right_inclusive=True, print_output=False, return_variable=True)
mltk.plot_variable_response(DataFrame=Data, variable=groupVariable, class_variable=targetVariable)

            Counts  HighIncome  CountsFraction%  ResponseFraction%  ResponseRate%
ageGRP                                                                           
1_(0,20]      2410           2          7.40149            0.02551        0.08299
2_(20,30]     8162         680         25.06680            8.67236        8.33129
3_(30,40]     8546        2406         26.24612           30.68486       28.15352
4_(40,50]     6983        2655         21.44590           33.86048       38.02091
5_(50,60]     4128        1547         12.67774           19.72963       37.47578
6_(60,INF)    2332         551          7.16194            7.02716       23.62779
TOTAL        32561        7841        100.00000          100.00000            NaN

# Create One Hot Encoded Variables
Data, featureVariables, targetVariable = mltk.to_one_hot_encode(Data, category_variables=categoryVariables, binary_variables=binaryVariables, target_variable=targetVariable)
Data[identifierColumns+featureVariables+[targetVariable]].sample(5).transpose()

Correlation

correlation=mltk.correlation_matrix(Data, featureVariables+[targetVariable], target_variable=targetVariable, method='pearson', return_type='list', show_plot=False)

Split Train, Validate Test datasets

TrainDataset, ValidateDataset, TestDataset = mltk.train_validate_test_split(Data, ratios=(0.6,0.2,0.2))

Model Building

sample_attributes = {'SampleDescription':'Adult Census Income Dataset',
                    'NumClasses':2,
                    'RecordIdentifiers':identifierColumns
                    }

score_parameters = {'Edges':[0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
                   'Quantiles':10,
                   'ScoreVariable':'Probability',
                   'ScoreLabel':'Score',
                   'QuantileLabel':'Quantile'
                   }

model_attributes = {'ModelID': None,   
                   'ModelName': 'IncomeLevel',
                   'Version':'0.1',
                   }

Losgistic Regression

model_parameters = {'MLAlgorithm':'LGR', # 'RF', # 'DFF', # 'CNN', # 'CATBST', # 'XGBST'
                    'MaxIterations':50}  

Random Forest

model_parameters = {'MLAlgorithm':'RF', # 'LGR', #  'DFF', # 'CNN', # 'CATBST', # 'XGBST'
                    'NTrees':500,
                   'MaxDepth':100,
                   'MinSamplesToSplit':10,
                   'Processors':2} 

Neural Networks

# Setup Architecture
# Binary classification (L1 'units': 2), 32 variables ('input_shape':(48,))
SimpleDFF_architecture = {
        'L1':{'type': 'Dense', 'position':'input', 'units': 512, 'activation':'relu', 'input_shape':(48,)},
        'L2':{'type': 'Dense', 'position':'hidden', 'units': 512, 'activation':'relu'},
        'L3':{'type': 'Dropout', 'position':'hidden', 'rate':0.5},
        'L4':{'type': 'Dense', 'position':'output', 'units': 2, 'activation':'softmax', 'output_shape':None},
       }

# Binary classification (L1 'units': 2), 32 variables ('input_shape':(32,))
LogisticRegressionNN_architecture = {
        'L1':{'type': 'Dense', 'position':'input', 'units': 2, 'activation':'softmax', 'input_shape':(32,)},
       }

# Binary classification (L8 'units': 2)
SimpleImageClassifier_architecture = {
        'L1':{'type': 'Conv2D', 'position':'input', 'filters': 32, 'kernel_size':(3,3), 'strides':(1,1), 'padding':'valid', 'activation':'relu', 'input_shape':(128, 128, 1)},
        'L2':{'type': 'Conv2D', 'position':'hidden', 'filters': 64, 'kernel_size':(3,3), 'strides':(1,1), 'padding':'valid', 'activation':'relu'},
        'L3':{'type': 'MaxPooling2D', 'position':'hidden', 'pool_size': (2,2), 'padding':'valid'},   
        'L4':{'type': 'Dropout', 'position':'hidden', 'rate':0.25},
        'L5':{'type': 'Flatten', 'position':'hidden'},        
        'L6':{'type': 'Dense', 'position':'hidden', 'units': 128, 'activation':'relu'},
        'L7':{'type': 'Dropout', 'position':'hidden', 'rate':0.5},
        'L8':{'type': 'Dense', 'position':'output', 'units': 2, 'activation':'softmax', 'output_shape':None},
       }

model_parameters = {'MLAlgorithm':'NN',
                    'BatchSize':512,
                   'InputShape':InputShape,
                   'num_classes':2,
                   'Epochs':10,
                   'metrics':['accuracy'],
                   'architecture':SimpleDFF_architecture} 

CatBoost [avaiable in v0.1.5]

model_parameters = {'MLAlgorithm':'CBST',
                    'NTrees': 500,
                    'MaxDepth':10,
                    'LearningRate':0.7,
                    'LossFunction':'Logloss',#crossEntropy
                    'EvalMatrics':'Accuracy',
                    'Imbalanced':False,
                    'TaskType':'GPU',
                    'Processors':2,
                    'UseBestModel':True}

CBSTModel = mltk.build_ml_model(TrainDataset, ValidateDataset, TestDataset, model_variables, targetVariable, 
                 model_attributes, sample_attributes, model_parameters, score_parameters, 
                          return_model_object=True, show_results=False, show_plot=True)

Build Model

RFModel = mltk.build_ml_model(TrainDataset, ValidateDataset, TestDataset, model_variables, targetVariable, 
                 model_attributes, sample_attributes, model_parameters, score_parameters, 
                          return_model_object=True, show_results=False, show_plot=True)

print(RFModel.model_attributes['ModelID'])
print(RFModel.model_interpretation['ModelSummary'])
print(RFModel.model_evaluation['AUC'])
print(RFModel.model_evaluation['RobustnessTable'])

RFModel.plot_eval_matrics(comparison=False)

Save model

saveFilePath = '{}.pkl'.format(RFModel.get_model_id())
mltk.save_model(RFModel, saveFilePath)

          minProbability  maxProbability  meanProbability  BucketCount  ResponseCount  BucketFraction  ResponseFraction  BucketPrecision  CumulativeBucketFraction  CumulativeResponseFraction  CumulativePrecision
Quantile                                                                                                                                                                                                           
1                0.00000         0.00008      3.85729e-06          652            3         0.10011           0.00192          0.00460                   1.00000                     1.00000              0.23967
2                0.00008         0.00432      1.52655e-03          651            9         0.09995           0.00577          0.01382                   0.89989                     0.99808              0.26582
3                0.00435         0.02042      1.10941e-02          652           14         0.10011           0.00897          0.02147                   0.79994                     0.99231              0.29731
4                0.02049         0.05702      3.58648e-02          650           20         0.09980           0.01281          0.03077                   0.69983                     0.98334              0.33677
5                0.05711         0.12075      8.51409e-02          652           65         0.10011           0.04164          0.09969                   0.60003                     0.97053              0.38767
6                0.12086         0.20457      1.63366e-01          651          109         0.09995           0.06983          0.16743                   0.49992                     0.92889              0.44533
7                0.20469         0.31870      2.61577e-01          651          190         0.09995           0.12172          0.29186                   0.39997                     0.85906              0.51478
8                0.31895         0.46840      4.03550e-01          666          259         0.10226           0.16592          0.38889                   0.30002                     0.73735              0.58905
9                0.46854         0.66965      5.68083e-01          641          377         0.09842           0.24151          0.58814                   0.19776                     0.57143              0.69255
10               0.66994         0.99967      8.06834e-01          647          515         0.09934           0.32992          0.79598                   0.09934                     0.32992              0.79598
DataSet          0.00000         0.99967      2.33167e-01         6513         1561         1.00000           1.00000          0.23967                   1.00000                     1.00000              0.23967

Evaluate Model

# Plot model performance curves
RFModel.plot_eval_matrics(comparison=True)
LGRModel.plot_eval_matrics(comparison=True)
NNModel.plot_eval_matrics(comparison=True)
CBSTModel.plot_eval_matrics(comparison=True)

TestDataset = mltk.score_processed_dataset(TestDataset, RFModel, edges=None, score_label=None, fill_missing=0)
score_variable = RFModel.get_score_variable()
score_label = RFModel.get_score_label()

Robustnesstable = mltk.robustness_table(ResultsSet=TestDataset, class_variable=targetVariable, score_variable=score_variable,  score_label=score_label, show_plot=True)

threshold = 0.8
TestDataset = mltk.set_predicted_columns(TestDataset, score_variable, threshold=threshold)
ConfusionMatrix = mltk.confusion_matrix(TestDataset, actual_variable=targetVariable, predcted_variable='Predicted', labels=[0,1], sample_weight=None, totals=True)
print(ConfusionMatrix)

Compare AUC [avaiable in v0.1.5]

Models = [LGRModel, RFModel, CBSTModel, NNModel]
ModelsComp = mltk.model_guages_comparison(Models)
ModelsComp.style.background_gradient(cmap='RdYlGn').set_precision(3)

Comparing Models and Probability Thresholds

Models = [LGRModel, RFModel, CBSTModel, NNModel]
thresholds=[0.7, 0.8, 0.9]
ConfusionMatrixComparison = mltk.confusion_matrix_comparison(TestDataset, Models, thresholds, score_variable=None, show_plot=True)
ConfusionMatrixComparison.style.background_gradient(cmap='RdYlGn').set_precision(3)

Comparing Models and Threshold Score (1-10 Scale) [avaiable in v0.1.5]

Models = [LGRModel, RFModel, CBSTModel, NNModel]
thresholds=[7, 8, 9]
ConfusionMatrixComparison = mltk.confusion_matrix_comparison(TestDataset, Models, thresholds, score_variable=score_label, show_plot=True)
ConfusionMatrixComparison.style.background_gradient(cmap='RdYlGn').set_precision(3)

Deployment

Simplified MLToolkit ETL pipeline for scoring and model re-building (Need to customize based on the project).

Define ETL Function

def ETL(DataFrame):
    # Add ID column
    DataFrame = mltk.add_identity_column(DataFrame, id_label='ID', start=1, increment=1)

    # Clean column names
    DataFrame = mltk.clean_column_names(DataFrame, replace='')
    input_columns = list(DataFrame.columns)

    # Excising valiables to model
    targetVariable = 'HighIncome'
    categoryVariables = ['sex', 'nativecountry', 'race', 'occupation', 'workclass', 'maritalstatus', 'relationship']
    binaryVariables = []

    # List to Create more Binary variables
    conditions = [{'bin_variable':'CapitalGainPositive', 'str_condition':"capitalgain>0"},
                  {'bin_variable':'CapitalLossPositive', 'str_condition':"capitalloss>0"}
                  ]

    # List to Create more Catergorical variables
    buckets = [{'variable':'age', 'str_labels':['0', '20', '30', '40', '50', '60', 'INF']},
               {'variable':'educationnum', 'str_labels':['1', '4', '6', '8', '10', '13', '16']},
               {'variable':'hoursperweek', 'str_labels':['0', '20', '40', '50', '60', '80', 'INF']}
              ] 

    # List to Merge categorical values
    groups = [{'variable':'maritalstatus', 'group_name':'Married', 'values':["Married-civ-spouse", "Married-spouse-absent", "Married-AF-spouse"]}]

    DataFrame, categoryVariables, binaryVariables, targetVariable = mltk.setup_variables(DataFrame, target_variable=targetVariable, category_variables=categoryVariables, binary_variables=binaryVariables, conditions=conditions, buckets=buckets, groups=groups)

    # Create One Hot Encoded Variables
    DataFrame, featureVariables, targetVariable = mltk.to_one_hot_encode(DataFrame, category_variables=categoryVariables, binary_variables=binaryVariables, target_variable=targetVariable)

    return DataFrame, input_columns

Scoring

MLModelObject = mltk.load_model(saveFilePath)
SampleDataset = pd.read_csv(r'test.csv')
SampleDataset = ETL(SampleDataset)

SampleDataset = mltk.score_processed_dataset(SampleDataset, MLModelObject, edges=None, score_label=None, fill_missing=0)
Robustnesstable1 = mltk.robustness_table(ResultsSet=SampleDataset, class_variable=targetVariable, score_variable=score_variable,  score_label=score_label, show_plot=True)

MLModelObject = mltk.load_model(saveFilePath)

TestInput = """
{
      "ID": "A001",
      "age": 32,
      "workclass": "Private",
      "education": "Doctorate",
      "education-num": 16,
      "marital-status": "Married-civ-spouse",
      "occupation": "Prof-specialty",
      "relationship": "Husband",
      "race": "Asian-Pac-Islander",
      "sex": "Feale",
      "capital-gain": 0,
      "capital-loss": 0,
      "hours-per-week": 40,
      "native-country": "?"
}
"""
output = mltk.score_records(TestInput, MLModelObject, edges=None, ETL=ETL, return_type='dict') # Other options for return_type, {'json', 'frame'}

Output

[{'ID': 'A001',
  'age': 32,
  'capitalgain': 0,
  'capitalloss': 0,
  'education': 'Doctorate',
  'educationnum': 16,
  'hoursperweek': 40,
  'maritalstatus': 'Married',
  'nativecountry': '?',
  'occupation': 'Prof-specialty',
  'race': 'Asian-Pac-Islander',
  'relationship': 'Husband',
  'sex': 'Feale',
  'workclass': 'Private',
  'Probability': 0.7494862543903595,
  'Score': 8}]

JSON Input for scoring

Records Format for single or fewer number of records

[{
	"ID": "A001",
	"age": 32,
	"workclass": "Private",
	"education": "Doctorate",
	"occupation": "Prof-specialty",
	"sex": "Female",
	"hoursperweek": 40,
	"nativecountry": "USA"
},]

Split Format for mulltiple records

{
	"columns":["ID","age","education","hoursperweek","nativecountry","occupation","sex","workclass"],
	"data":[["A001",32,"Doctorate",40,"USA","Prof-specialty","Female","Private"],]
}

License

Copyright 2019 Sumudu Tennakoon

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

MLToolkit Project Timeline

• 2018-07-02 [v0.0.1]: Initial set of functions for data exploration, model building and model evaluation was published to Github. (https://github.com/sptennak/MachineLearning).
• 2018-01-03 [v0.0.2]: Created more functions for data exploration including web scraping and geo spacial data analysis for for IBM Coursera Data Science Capstone Project was published to Github. (https://github.com/sptennak/Coursera_Capstone).
• 2019-03-20 [v0.1.0]: Developed and published initial version of model building and serving framework for IBM Coursera Advanced Data Science Professional Certificate Capstone Project. (https://github.com/sptennak/IBM-Coursera-Advanced-Data-Science-Capstone).
• 2019-07-02 [v0.1.2]: First release of the PyMLToolkit Python package, a collection of clases and functions facilitating end-to-end machine learning model building and serving over RESTful API.
• 2019-07-04 [v0.1.3]: Minor bug fixes.
• 2019-07-14 [v0.1.4]: Improved documentation, Integrated TensorFlow Models, Enhancements and Minor bug fixes.
• 2019-07-20 [v0.1.5(preview)]: Improved documentation, Integrated CatBoost Models, String Operations, Major Enhancements and Minor bug fixes.

Future Release Plan

• 2019-08-XX [v0.1.5]: Integrate CatBoost Models, , String Operations, Integrate image classification model Deployment, Enhancements and Minor bug fixes.
• 2019-12-31 [v0.1.6]: Comprehensive documentation, Major bug-fix version of the initial release with some enhancements.
• [v0.2.0]: Imporved model serving frameework, support more machine learning algorithms and deep learning.
• [v0.3.0]: Imporved scalability and performance, Hyper parameter tuning, and Automated Machine Learning.
• [v0.4.0]: Building continious learning models.

References

• https://pandas.pydata.org/
• https://scikit-learn.org
• https://www.numpy.org/
• https://docs.python.org/3.6/library/re.html
• https://www.statsmodels.org
• https://matplotlib.org/
• http://flask.pocoo.org/