A starter implementation for logistic regression.

logistic_regression_example.py

import pandas as pd
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import confusion_matrix
from sklearn.model_selection import train_test_split

def load_data(path):
  # Load the data into a Pandas dataframe.
  raw_data = pd.read_csv(path, header=0)  
  print('Loaded data from', path)
  return raw_data

def show_samples(data, num_samples=5):
    print(raw_data[:num_samples])
    print('')
    print('Count by target feature:')
    print(raw_data[TARGET_FEATURE].value_counts())
    print('')
    
def get_usable_data(data, add_intercept=True, categorical_features=[]):
  # Separate the X and Y values.
  y_data = raw_data[TARGET_FEATURE]
  x_data = raw_data.drop(TARGET_FEATURE, axis=1)

  # To include an intercept, add a new column with a constant.
  if add_intercept:
    x_data['intercept'] = 1.0

  for feature in categorical_features:
    dummy_data = pd.get_dummies(x_data[feature], prefix=feature)

    # We need to remove at least one of the dummy features.
    # It doesn't matter which one. To stay consistent we
    # usually remove the most common value.
    most_common_value = pd.value_counts(x_data[feature]).index[0]
    dummy_to_exclude = feature + '_' + str(most_common_value)
    dummy_data_to_use = dummy_data.drop(dummy_to_exclude, axis=1)
    x_data[dummy_data_to_use.columns] = dummy_data_to_use

    # Remove the original feature, so just the dummy data remains.
    x_data = x_data.drop(feature, axis=1)
  
  return x_data, y_data

def evaluate(model, x_test):
  # Get prediction probabilities for the test set.
  y_predict_proba_raw = model.predict_proba(x_test)

  # The .predict_proba() function returns the probabilities of Y
  # equaling both 1 and 0, so we just want the probability that
  # Y equals 1.
  y_predict_proba = y_predict_proba_raw[:, 1]

  # Scale the probabilities to predictions of 1 or 0, based on
  # a threshold of 50%.
  y_predict = [0 if y < 0.5 else 1 for y in y_predict_proba]
 
def get_confusion_matrix_stats(y_test, y_predict):
  # Get the confusion matrix and calculate the results.
  matrix = confusion_matrix(y_test, y_predict)
  n_samples = float(len(y_test))
  accuracy = float(matrix[0][0] + matrix[1][1]) / n_samples
  precision = matrix[1][1] / float(matrix[0][1] + matrix[1][1])
  recall    = matrix[1][1] / float(matrix[1][0] + matrix[1][1])
  print('Accuracy: %.2f'  % accuracy)
  print('Precision: %.2f' % precision)
  print('Recall: %.2f'    % recall)


# Usage:

# The target feature is whether or not the employee left.
TARGET_FEATURE = 'left'
TEST_SET_SIZE = 0.2
CATEGORICAL_FEATURES = ['sales', 'salary']

raw_data = load_data('./HR_comma_sep.csv')
# It's helpful to take a quick look at the data.
show_samples(raw_data)

# Preprocess
x_data, y_data = get_usable_data(raw_data, categorical_features=CATEGORICAL_FEATURES)
# Good to eye-ball the processed data too
show_samples(raw_data)

x_train, x_test, y_train, y_test = train_test_split(x_data, y_data, test_size=TEST_SET_SIZE)
model = LogisticRegression().fit(x_train, y_train)
predictions = evaluate(model, x_test)
get_confusion_matrix(y_test, predictions)