Merge pull request #7 from matthewgillett/bias-variance

Bias-variance decomposition feature

Author: matthewgillett

.pre-commit-config.yaml

@@ -9,20 +9,20 @@
 
 repos:
 
-- repo: https://github.com/ambv/black
-  rev: 22.3.0
+- repo: https://github.com/psf/black
+  rev: 23.11.0
   hooks:
   - id: black
     language_version: python3.8
     args: [--line-length=88, tests, mvtk]
 
-- repo: https://gitlab.com/pycqa/flake8
-  rev: 4.0.1
+- repo: https://github.com/pycqa/flake8
+  rev: 6.1.0
   hooks:
   - id: flake8
     args: [--max-line-length=88, '--per-file-ignores=__init__.py:F401,F403', tests, mvtk]
 - repo: https://github.com/pre-commit/mirrors-mypy
-  rev: v0.942
+  rev: v1.7.1
   hooks:
   - id: mypy
     files: mvtk/

README.md

@@ -32,6 +32,7 @@ You can import:
 - `mvtk.sobol` for Sobol sensitivity analysis
 - `mvtk.supervisor` for divergence analysis
 - `mvtk.metrics` for specialised metrics
+- `mvtk.bias_variance` for bias-variance decomposition
 
 # Documentation
 You can run `make -C docs html` on a Mac or `make.bat -C docs html` on a PC to just rebuild the docs. In this case, point your browser to ```docs/_build/html/index.html``` to view the homepage. If your browser was already pointing to documentation that you changed, you can refresh the page to see the changes.

docs/bias_variance.bias_variance.rst

@@ -0,0 +1,8 @@
+bias_variance
+=============
+
+.. automodule:: mvtk.bias_variance.bias_variance
+    :members:
+    :imported-members:
+    :undoc-members:
+    :show-inheritance:

docs/bias_variance.bias_variance_parallel.rst

@@ -0,0 +1,8 @@
+bias_variance_parallel
+======================
+
+.. automodule:: mvtk.bias_variance.bias_variance_parallel
+    :members:
+    :imported-members:
+    :undoc-members:
+    :show-inheritance:

docs/bias_variance.estimators.rst

@@ -0,0 +1,8 @@
+estimators
+==========
+
+.. automodule:: mvtk.bias_variance.estimators
+    :members:
+    :imported-members:
+    :undoc-members:
+    :special-members: __init__, __call__

docs/bias_variance.rst

@@ -0,0 +1,11 @@
+bias_variance
+=============
+
+Subpackages
+-----------
+
+.. toctree::
+
+    bias_variance.estimators
+    bias_variance.bias_variance
+    bias_variance.bias_variance_parallel

docs/bias_variance_user_guide.rst

@@ -0,0 +1,225 @@
+########################
+Bias-Variance User Guide
+########################
+
+**********
+Motivation
+**********
+
+Statistical Bias vs. "Fairness"
+===============================
+
+For this user guide and associated submodule, we are referring to
+`statistical bias <https://en.wikipedia.org/wiki/Bias_(statistics)>`_ rather
+than the "fairness" type of bias.
+
+Why should we care about bias and variance?
+===========================================
+
+Bias and variance are two indicators of model performance and together represent
+two-thirds of model error (the remaining one-third is irreducible "noise" error that
+comes from the data set itself). We can define bias and variance as follows
+by training a model with multiple `bootstrap sampled
+<https://en.wikipedia.org/wiki/Bootstrapping_(statistics)>`_ training sets, resulting in
+multiple instances of the model.
+
+.. topic:: Bias and variance defined over multiple training sets:
+
+    * Bias represents the average difference between the prediction a model makes and the correct prediction.
+    * Variance represents the average variability of the prediction a model makes.
+
+Typically, a model with high bias is "underfit" and a model with high variance is
+"overfit," but keep in mind this is not always the case and there can be many reasons
+why a model has high bias or high variance. An "underfit" model is oversimplified and
+performs poorly on the training data, whereas an "overfit" model sticks too closely to
+the training data and performs poorly on unseen examples. See Scikit-Learn's
+`Underfitting vs. Overfitting
+<https://scikit-learn.org/stable/auto_examples/model_selection/plot_underfitting_overfitting.html>`_
+for a clear example of an "underfit" model vs. an "overfit" model.
+
+There is a concept
+known as the `"bias-variance tradeoff"
+<https://en.wikipedia.org/wiki/Bias%E2%80%93variance_tradeoff>`_ that describes
+the relationship between high bias and high variance in a model. Our ultimate goal
+here is to find the ideal balance where both bias and variance is at a minimum.
+It is also important from a business problem standpoint on whether the model
+error that we are unable to reduce should favor bias or variance.
+
+*****************************************
+Visualize Bias and Variance With Examples
+*****************************************
+
+In order to easily understand the concepts of bias and variance, we will show
+four different examples of models for each of the high and low bias and variance
+combinations. These are extreme and engineered cases for the purpose of clearly
+seeing the bias/variance.
+
+Before we begin, let's take a look at the distribution of the labels. Notice
+that the majority of label values are around 1 and 2, and much less around 5.
+
+.. figure:: images/bias_variance_label_distribution.png
+    :align: center
+    :alt: alternate text
+    :figclass: align-center
+
+First we have a model with high bias and low variance. We artificially
+introduce bias to the model by adding 10 to every training label, but leaving
+the test labels as is. Given that values of greater than 5 in the entire label
+set are considered outliers, we are fitting the model against outliers.
+
+.. figure:: images/high_bias_low_variance.png
+    :align: center
+    :alt: alternate text
+    :figclass: align-center
+
+    Five sets of mean squared error results from the test set from the five
+    bootstrap sample trainings of the model. Notice the model error is very
+    consistent among the trials and is not centered around 0.
+
+Next we have a model with low bias and high variance. We simulate this by
+introducing 8 random "noise" features to the data set. We also reduce the size
+of the training set and train a neural network over a low number of epochs.
+
+.. figure:: images/low_bias_high_variance.png
+    :align: center
+    :alt: alternate text
+    :figclass: align-center
+
+    Five sets of mean squared error results from the test set from the five
+    bootstrap sample trainings of the model. Notice the model error has
+    different distributions among the trials and centers mainly around 0.
+
+Next we have a model with high bias and high variance. We simulate through
+a combination of the techniques from the high bias low variance example and
+the low bias high variance example and train another neural network.
+
+.. figure:: images/high_bias_high_variance.png
+    :align: center
+    :alt: alternate text
+    :figclass: align-center
+
+    Five sets of mean squared error results from the test set from the five
+    bootstrap sample trainings of the model. Notice the model error has
+    different distributions among the trials and is not centered around 0.
+
+Finally we have a model with low bias and low variance. This is a simple
+linear regression model with no modifications to the training or test labels.
+
+.. figure:: images/low_bias_low_variance.png
+    :align: center
+    :alt: alternate text
+    :figclass: align-center
+
+    Five sets of mean squared error results from the test set from the five
+    bootstrap sample trainings of the model. Notice the model error is very
+    consistent among the trials and centers mainly around 0.
+
+***************************
+Bias-Variance Decomposition
+***************************
+
+.. currentmodule:: mvtk.bias_variance
+
+There are formulas for breaking down total model error into three parts: bias,
+variance, and noise. This can be applied to both regression problem loss
+functions (mean squared error) and classification problem loss functions
+(0-1 loss). In a paper by Pedro Domingos, a method of unified
+decomposition was proposed for both types of problems :cite:`domingos2000decomp`.
+
+First lets define :math:`y` as a single prediction, :math:`D` as the set of
+training sets used to train the models, :math:`Y` as the set of predictions
+from the models trained on :math:`D`, and a loss function :math:`L` that
+calculates the error between our prediction :math:`y` and the correct
+prediction.
+The main prediction :math:`y_m` is the smallest average loss for a prediction
+when compared to the set of predictions :math:`Y`. The main prediction is
+the mean of :math:`Y` for mean squared error and the mode of :math:`Y` for
+0-1 loss :cite:`domingos2000decomp`.
+
+Bias can now be defined for a single example :math:`x` over the set of models
+trained on :math:`D` as the loss calculated between the main prediction
+:math:`y_m` and the correct prediction :math:`y_*` :cite:`domingos2000decomp`.
+
+.. math::
+    B(x) = L(y_*,y_m)
+
+Variance can now be defined for a single example :math:`x` over the set of
+models trained on :math:`D` as the average loss calculated between all predictions
+and the main prediction :math:`y_m` :cite:`domingos2000decomp`.
+
+.. math::
+    V(x) = E_D[L(y_m, y)]
+
+We will need to take the average of the bias over all examples as
+:math:`E_x[B(x)]` and the average of the variance over all examples as
+:math:`E_x[V(x)]` :cite:`domingos2000decomp`.
+
+With :math:`N(x)` representing the irreducible error from observation noise, we
+can decompose the average expected loss as :cite:`domingos2000decomp`
+
+.. math::
+    E_x[N(x)] + E_x[B(x)] + E_x[cV(x)]
+
+In other words, the average loss over all examples is equal to the noise plus the
+average bias plus the net variance (the :math:`c` factor included with the variance
+when calculating average variance gives us the net variance).
+
+.. note::
+    We are generalizing the actual value of :math:`N(x)`, as the Model Validation
+    Toolkit's implementation of bias-variance decomposition does not include noise
+    in the average expected loss. This noise represents error in the actual data
+    and not error related to the model itself. If you would like to dive deeper
+    into the noise representation, please consult the `Pedro Domingos paper
+    <https://homes.cs.washington.edu/~pedrod/papers/mlc00a.pdf>`_.
+
+For mean squared loss functions, :math:`c = 1`, meaning that average variance
+is equal to net variance.
+
+For zero-one loss functions, :math:`c = 1` when :math:`y_m = y_*` otherwise
+:math:`c = -P_D(y = y_* | y != y_m)`. :cite:`domingos2000decomp` In other words,
+:math:`c` is 1 when the main prediction is the correct prediction. If the main
+prediction is not the correct prediction, then :math:`c` is equal to the
+probability of a single prediction being the correct prediction given that the
+single prediction is not the main prediction.
+
+Usage
+=====
+
+:meth:`bias_variance_compute` will return the average loss, average bias, average
+variance, and net variance for an estimator trained and tested over a specified number
+of training sets. This was inspired and modeled after Sebastian Raschka's
+`bias_variance_decomp
+<https://github.com/rasbt/mlxtend/blob/master/mlxtend/evaluate/bias_variance_decomp.py>`_
+function :cite:`mlxtenddecomp`.
+We use the `bootstrapping <https://en.wikipedia.org/wiki/Bootstrapping_(statistics)>`_
+method to produce our sets of training data from the original training set. By default
+it will use mean squared error as the loss function, but it will accept the following
+functions for calculating loss.
+
+* :meth:`bias_variance_mse` for mean squared error
+* :meth:`bias_variance_0_1_loss` for 0-1 loss
+
+Since :meth:`bias_variance_compute` trains an estimator over multiple iterations, it also
+expects the estimator to be wrapped in a class that extends the
+:class:`estimators.EstimatorWrapper` class, which provides fit and predict methods
+that not all estimator implementations conform to. The following estimator wrappers are
+provided.
+
+* :class:`estimators.PyTorchEstimatorWrapper` for `PyTorch <https://pytorch.org/>`_
+* :class:`estimators.SciKitLearnEstimatorWrapper` for `Scikit-Learn <https://scikit-learn.org/stable/>`_
+* :class:`estimators.TensorFlowEstimatorWrapper` for `TensorFlow <https://www.tensorflow.org/>`_
+
+:meth:`bias_variance_compute` works well for smaller data sets and less complex models, but what
+happens when you have a very large set of data, a very complex model, or both?
+:meth:`bias_variance_compute_parallel` does the same calculation, but leverages `Ray
+<https://www.ray.io/>`_ for parallelization of bootstrapping, training, and predicting.
+This allows for faster calculations using computations over a distributed architecture.
+
+.. topic:: Tutorials:
+
+    * :doc:`Bias-Variance Visualization <notebooks/bias_variance/BiasVarianceVisualization>`
+    * :doc:`Bias-Variance Regression <notebooks/bias_variance/BiasVarianceRegression>`
+    * :doc:`Bias-Variance Classification <notebooks/bias_variance/BiasVarianceClassification>`
+
+.. bibliography:: refs.bib
+    :cited:

docs/conf.py

@@ -23,7 +23,7 @@
 author = "Model Validation Toolkit Team"
 
 # The full version, including alpha/beta/rc tags
-release = "0.0.1"
+release = "0.2.0"
 
 
 # -- General configuration ---------------------------------------------------
@@ -73,8 +73,8 @@
 # A fix for Sphinx error contents.rst not found
 master_doc = "index"
 
-# increase the timeout for long running notebooks
-nbsphinx_timeout = 180
+# increase the timeout for long-running notebooks
+nbsphinx_timeout = 900
 
 # Don't show full paths
 add_module_names = False