utils

utils

contains different kind of helper functions and classes

---

_helper.py

implements abstract classes as support for the other algorithms - for example implementing the score() function for all classifiers and regressors.

Both possible use cases for Machine Learning algorithms are implemented, the abstract implementation of a standard Classifier and a standard Regressor, both used for Supervised Learning. Similar to sklearn the following functionalities are implemented

• _classifier - the standart Classifier with
  • train()
  • predict()
  • score()
• _regressor - the standard Regressor with
  • train()
  • predict()
  • score()

Furthermore added functionality:

• convertSeconds() that converts seconds into hours, minutes and seconds

---

metrics.py

implements several loss-/activation-functions (for deep_learning) and the metrics for score()-evaluation

The following functionality is implemented here:

• get_activation_function() providing different kinds of activation functions for the Layers of a Neural Networks. A short description about activation functions can be found here. The following ones are implemented:

  • Sigmoid
  • Tangens Hyperbolicus
  • Rectified Linear Unit
  • Leaky Rectified Linear Unit
  • Soft-Max

• loss_function() provides different kind of loss functions used for the training of a Neural Network. A short introduction towards loss functions can be found here. The following of the presented functions are implemented:

  • Mean squared Error for Regression
  • Mean absolute Error for Regression
  • Mean squared logarithmic Error for Regression
  • Hinge for binary classification
  • Squared Hinge for binary classification
  • Cross Entropy for binary classification - here called Binary Cross Entropy
  • Categorical Cross Entropy for multi-class classification - here called Multi-Class Cross Entropy
  • Kullback-Leibler Divergence for multi-class classification

  Furthermore added are:

  • L1-norm Error for Regression. Same as Mean absoulte Error without normalizing
  • L2-norm Error for Regression. Same as Mean squared Error without normalizing
  • Root mean squared Error for Regression
  • Huber for Regression

• calc_rates() calculates the true positives, the false positives, the true negatives and the false negatives each class → What are these rates?

• classifier_score() calculates possible evaluation metrics - described here - for a Classifier, including:

  • Recall
  • Precision
  • Accuracy
  • F1 score
  • Balanced Accuracy being similar to Accuracy but paying respect to an unbalanced number of samples for different classes

• regressor_score() calculates possible evaluation metrics - described above - for a Regressor, including:

  • L1-norm
  • L2-norm
  • Mean squared Error
  • Mean absolute Error
  • Root mean squared Error

---

preprocessing.py

implements classes and functions for data preprocessing:

• MinMaxScaler - Splits arrays or matrices into random train and test subsets, similar to MinMaxScaler by sklearn. Here you will find a brief introduction into the meaning of Scaling, Standardizing and Normalizing
• train_test_split - implements a MinMaxScaler that scales all given data into the range from 0 to 1, similar to train_test_split by sklearn. Why splitting your dataset into train and test samples is important, can be found here.