DVHA-Machine-Learning-Core

Core library of machine learning visualizations developed for DVH Analytics, adapted for generic use. Under active development.

Simply provide X and y arrays and get a GUI for data exploration and visualization. This application handles data splitting and modeling for you. Algorithms based on scikit-learn.

The code is built upon these core libraries:

• DVH Analytics - A DICOM Database Application for Radiation Oncology
• wxPython Phoenix - Build a native GUI on Windows, Mac, or Unix systems
• Bokeh - Interactive Web Plotting for Python
• scikit-learn - Machine Learning in Python

To Run

NOTE: This application supports python >= 3.5

Either clone this project or install from PyPI:
pip install dvha-mlc

From a python3 console:

from dvhamlc.gui import Model
ModelApp(X, y)

Optionally specify an algorithm:

ModelApp(X, y, algorithm='random_forest', predictive_type='regression')

class ModelApp(X, y, algorithm='random_forest', predictive_type='regression', y_variable='Dependent Variable', y_categories=None, x_variables=None)

• Parameters
  • X : array-like, shape (n_samples, n_features)
  • y : array-like, shape (n_samples)
  • algorithm : string, optional (default=’random_forest’)
    • Defines which machine learning algorithm to use. Currently supported values (and their sources) include:
      • random_forest
        • sklearn.ensemble.RandomForestRegressor
        • sklearn.ensemble.RandomForestClassifier
      • gradient_boosting
        • sklearn.ensemble.GradientBoostingRegressor
        • sklearn.ensemble.GradientBoostingClassifier
      • support_vector_machine
        • sklearn.svm.SVR
        • sklearn.svm.SVC
      • decision_tree
        • sklearn.tree.DecisionTreeRegressor
        • sklearn.tree.DecisionTreeClassifier
  • predictive_type : string, optional (default='regression')
    • May be either 'regression' or 'classification'
  • y_variable : string, optional (default='Dependent Variable')
    • Visual attribute only, sets the y-axis title
  • y_categories : array-like, shape (n_samples), optional (default=None)
    • Only applicable when predictive_type='classification'. This array maps the numerical values fed into the machine learning algorithm to labels, these values will appear in the hover messages upon plot inspection.
  • x_variables : array-like, shape (n_features), optional (default=None)
    • The feature importance plot uses this data to display variable names instead column indices

Dependencies

• Python >=3.5
• wxPython Phoenix >= 4.0.4
• Bokeh >= 1.2.0
• NumPy 1.16.4 tested
• Scikit-learn 0.21.2 tested

Visual Options

See options.py for visual customization of some sizes and colors

Example

Based on a Random Forest Classifier example by Chris Albon.

# Import the DVHA Machine Learning Core application
from dvhamlc.gui import ModelApp

# Load the library with the iris dataset
from sklearn.datasets import load_iris

# Load pandas and numpy
import pandas as pd
import numpy as np

# Set random seed
np.random.seed(0)

# Create an object called iris with the iris data
iris = load_iris()

# Create a dataframe with the four feature variables
df = pd.DataFrame(iris.data, columns=iris.feature_names)
# Get column of the species names, this is what we are going to try to predict
y = pd.Categorical.from_codes(iris.target, iris.target_names)

# Launch the DVHA Machine Learning Core application
ModelApp(df, y, predictive_type='classification')

TODO

• Design a view for cross-validation (i.e., load a model with new data, no new modeling)
• Store a history of models for comparison
• Methods for hyper-parameter grid search
• Add more analysis tools (e.g., confusion matrix)