src/diarc.py

# -*- coding: utf-8 -*-
"""diarc.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1Jyccp5Aeml-7oZABbACY2VTE9iQJg9Pe

# Bismillahir Rahmaanir Raheem
# Almadadh Ya Gause RadiAllahu Ta'alah Anh - Ameen

# <font color=grey>DIabetes-related Amputation Risk Calculator (DIARC)</font>
<b>_by Zakia Salod_</b>
"""

!pip install pycaret

from pycaret.utils import version
version()

from pycaret.utils import enable_colab
enable_colab()

import numpy as np # Linear algebra
import pandas as pd # Data processing, CSV file I/O (e.g. pd.read_csv)
import matplotlib.pyplot as plt # For graphical representations of the data
import seaborn as sns

# Just to make sure the results are reproducible
np.random.seed(1234)

dataset = pd.read_excel('amputation_dataset.xlsx')

print(dataset['AMPUTATION'].value_counts())

ax = sns.countplot(x="AMPUTATION", data=dataset)

# show the number of duplicate rows in this dataset
dataset.duplicated(keep='first').sum()

# remove the duplicate rows in this dataset
# only keep the first instance of the row
dataset = dataset.drop_duplicates(keep='first')

print(dataset['AMPUTATION'].value_counts())

ax = sns.countplot(x="AMPUTATION", data=dataset)

dataset.head()

# Under sample the dataset to handle the imbalance
# Shuffle the Dataset.
shuffled_dataset = dataset.sample(frac=1, random_state=4)

# Put all the amputation class in a separate dataset.
amputation_dataset = shuffled_dataset.loc[shuffled_dataset['AMPUTATION'] == 1]


#Randomly select 105 observations from the non-amputation (majority class)
non_amputation_dataset = shuffled_dataset.loc[shuffled_dataset['AMPUTATION'] == 0].sample(n=105,random_state=42)

# Concatenate both dataframes again
dataset = pd.concat([amputation_dataset, non_amputation_dataset])

print(dataset['AMPUTATION'].value_counts())

ax = sns.countplot(x="AMPUTATION", data=dataset)

dataset.to_excel('amputation_removed_duplicates_and_balanced.xlsx')

from pycaret.classification import *

clf = setup(data = dataset, target = 'AMPUTATION',  session_id = 42)

# display the dataset (X_train)
get_config('X_train')
# converts age from numeric to float
# converts gender and diabetes_class (the two binary category variables) into label encoder conversion
# so, gender_f ---> with value 1 indicating FEMALE is TRUE and value 0 indicating FEMALE is FALSE (and instead, MALE)
# diabetes_class type 1 diabetes ---> value 1 indicates diabetes type 1 and value 0 means diabetes type 2
# then, one hot encoding is applied to the race column (each race is split into separate columns, with value 1 denoting TRUE for that race)

# display the dataset (y_train)
get_config('y_train')

best_model = compare_models(sort = 'AUC')

# BLEND MODELS, ALHUM
# create models for blending
nb = create_model('nb')
bagged_nb = ensemble_model(nb, method='Bagging')
lr = create_model('lr')
bagged_lr = ensemble_model(lr, method='Bagging')
lda = create_model('lda')
bagged_lda = ensemble_model(lda, method='Bagging')

rf = create_model('rf')
bagged_rf = ensemble_model(rf, method='Bagging')
ada = create_model('ada')
bagged_ada = ensemble_model(ada, method='Bagging')


blend_specific = blend_models(estimator_list = [bagged_nb, bagged_lr, bagged_lda, bagged_rf, bagged_ada])

# plot model
plot_model(blend_specific)

# tuning
tuned_blend_specific = tune_model(blend_specific)

evaluate_model(tuned_blend_specific)

tuned_blend_specific_predictions = predict_model(tuned_blend_specific)

# finalize model for deployment
final_tuned_blend_specific = finalize_model(tuned_blend_specific)

# save the model
# creates a .pkl file
save_model(tuned_blend_specific, "tuned_blend_specific_model_19112021", verbose=True)

# display the dataset (X_test)
get_config('X_test')

# display the dataset (y_test)
get_config('y_test')

dataset2 = pd.read_excel('amputation_removed_duplicates_and_balanced.xlsx')

!pip install pandas-profiling

from pandas_profiling import ProfileReport

profile = ProfileReport(dataset2, title="Pandas Profiling Report")

profile.to_file("amputation_removed_duplicates_and_balanced_report.html")