{
 "cells": [
  {
   "cell_type": "markdown",
   "source": [
    "# Overview of the base class structure\n",
    "\n",
    "`aeon` uses a core inheritance hierarchy of classes across the toolkit, with\n",
    "specialised sub classes in each module. The basic class hierarchy is summarised in\n",
    "the following simplified UML\n",
    "\n",
    "<img src=\"img/uml.png\" width=\"800\" alt=\"Basic class hierarchy\">\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## sklearn `BaseEstimator` and aeon `BaseObject`\n",
    "\n",
    "To make sense of this, we break it down from the top.\n",
    "\n",
    "<img src=\"img/uml1.png\" width=\"600\" alt=\"Top level class hierarchy\">\n",
    "\n",
    "Everything inherits from sklearns `BaseEstimator`. This handles the mechanisms for\n",
    "getting and setting parameters. The code structure below is stylised to show the\n",
    "main functionality and may differ in details from the actual implementations.\n",
    "\n",
    "<img src=\"img/sklearn_base.png\" width=\"500\" alt=\"sklearn base\">\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "The aeon class `BaseObject` extends `BaseEstimator` and adds the tagging method and\n",
    "some other functionality used in `aeon` estimators\n",
    "\n",
    "<img src=\"img/base_object.png\" width=\"600\" alt=\"Base object\">\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## aeons ``BaseEstimator``, `BaseDeepNetwork` and `BaseMetric`\n",
    "\n",
    "Three classes extend `BaseObject`: ``BaseEstimator``, `BaseDeepNetwork` and\n",
    "`BaseMetric`.\n",
    "\n",
    "`BaseDeepNetwork` is the base class for all the deep learning networks defined in the\n",
    "`networks` module. It has a single abstract method `build_network`.\n",
    "\n",
    "<img src=\"img/base_deep_network.png\" width=\"600\" alt=\"Base object\">\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "The `BaseMetric` class is the base class for forecasting performance metrics. It has\n",
    "a single abstract method `evaluate`.\n",
    "\n",
    "<img src=\"img/base_metric.png\" width=\"600\" alt=\"Base metric\">\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "The ``BaseEstimator`` class is the base class for the majority of classes in aeon.\n",
    "Anything that uses fit and predict in aeon. It contains a protected attribute\n",
    "`_is_fitted` and checks as to the value of this attribute. It also has a method to\n",
    "get fitted parameters.\n",
    "\n",
    "<img src=\"img/base_estimator.png\" width=\"500\" alt=\"Base Estimator\">"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "`BaseEstimator` has four direct base classes: `BaseForecaster`,\n",
    "`BaseSeriesAnnotator`, `BaseTransformer` and `BaseCollectionEstimator`.\n",
    "\n",
    "\n",
    "<img src=\"img/uml2.png\" width=\"600\" alt=\"Top level class hierarchy\">"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## `BaseForecaster` (aeon.forecasting.base)\n",
    "contains the forecasting specific methods. More details are available in the [API](https://www.aeon-toolkit.org/en/latest/api_reference/forecasting.html). `BaseForecaster` has the following concrete methods:\n",
    "\n",
    "<img src=\"img/base_forecaster.png\" width=\"700\" alt=\"Base forecaster\">"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## `BaseSeriesAnnotator` (aeon.annotation.base)\n",
    "\n",
    "is a largely experimental base class for use with annotators that implement\n",
    "techniques for segmentation and anomaly detection. It has two public attributes: `fmt`\n",
    " and `labels`, and public methods as follows:\n",
    "\n",
    "<img src=\"img/base_annotator.png\" width=\"700\" alt=\"Base annotator\">"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## `BaseTransformer` (aeon.transformations.base)\n",
    "\n",
    "Is the base class for all transformers, including single series transformers and\n",
    "collections transformers.\n",
    "\n",
    "<img src=\"img/base_transformer.png\" width=\"700\" alt=\"Base transformer\">\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## `BaseCollectionEstimator` (aeon.base)\n",
    "\n",
    "Is the base class for estimators that construct models on collections of time series.\n",
    " This includes classifiers, clusterers, regressors and collection transformers. It\n",
    " contains attributes and tags common to all these estimators, and protected methods\n",
    " to perform  checks and preprocessing common to all these estimators.\n",
    "\n",
    "<img src=\"img/base_collection_estimator.png\" width=\"700\" alt=\"Base\n",
    "transformer\">"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "The subclasses of `BaseCollectionEstimator` are as follows\n",
    "\n",
    "<img src=\"img/uml3.png\" width=\"600\" alt=\"Top level class hierarchy\">\n",
    "\n",
    "they have similar interfaces, but they are not identical\n",
    "\n",
    "## `BaseClassifier` (aeon.classification)\n",
    "\n",
    "This is the base class for all classifiers. It uses the standard `fit`, `predict` and\n",
    " `predict_proba` structure from `sklearn`. `fit` and\n",
    " `predict` call the abstract methods `_fit` and `_predict` which are implemented in\n",
    " the subclass to define the classification algorithm. All of the common format checking\n",
    " and conversion is done using the following final methods defined in\n",
    " `BaseCollectionEstimator`.\n",
    "\n",
    "<img src=\"img/base_classifier.png\" width=\"700\" alt=\"Top level class hierarchy\">"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## `BaseRegressor` (aeon.regression)\n",
    "\n",
    "BaseRegressor has the same structure as `BaseClassifier`, although it has no\n",
    "`predict_proba` method. The tests on y are also different.\n",
    "\n",
    "\n",
    "<img src=\"img/base_regressor.png\" width=\"700\" alt=\"Top level class hierarchy\">"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## `BaseClusterer` (aeon.clustering)\n",
    "\n",
    "`BaseClusterer` also has `fit` and `predict`, but does not take input y. It does\n",
    "include `predict_proba`.\n",
    "\n",
    "\n",
    "<img src=\"img/base_clusterer.png\" width=\"700\" alt=\"Base\n",
    "transformer\">"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## `BaseCollectionTransformer` (aeon.transformations.collection)\n",
    "\n",
    "The `BaseCollectionTransformer` was introduced to differentiate transformers that\n",
    "work on a single series to those that work on collections. Part of the motivation was\n",
    " to work around a lot of legacy code in `BaseTransformer` that performs a huge amount\n",
    "  of conversion checks that is unnecessary for collections. Rather than `fit` and\n",
    "  `predict` it implements `fit`, `transform` and `fit_transform`.\n",
    "\n",
    "\n",
    "<img src=\"img/base_collection_transformer.png\" width=\"700\" alt=\"Base\n",
    "transformer\">\n"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [],
   "metadata": {
    "collapsed": false
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}