.. _cross_validation:

Cross validation assessment of PC models
----------------------------------------

Once a polynomial response surface :math:`\widehat{h}(\underline{x})`
of the original numerical model :math:`h(\underline{x})` has been
built up, it is crucial to estimate the approximation error, i.e. the
discrepancy between the response surface and the true model response
in terms of a suitable norm such as the :math:`L_2`-norm:

.. math::
   Err \, \, = \, \, \left\| h(\underline{x}) \; - \; \widehat{h}(\underline{x}) \right\|_{L_2}^2\, \, = \, \, \int_{\cD} \; \left( h(\underline{x}) \; - \; \widehat{h}(\underline{x}) \right)^2  \; d\underline{x}

where :math:`\cD` denotes the support of the input parameters
:math:`\underline{x}`. It is worth emphasizing that one tends to
overestimate the performance of a response surface by training and
evaluating it on the same data set. For instance, the model might fail
to predict on new data whereas the validation on the training data
yields satisfactory performance. To avoid such issue, it is important
that the model error assessment is conducted on a data set which is
independent of the training sample. However, any new model evaluation
may be time- and memory-consuming. Therefore, error estimates which
are only based on already performed computer experiments are of
interest. In this context, the so-called *cross validation* techniques
are utilized to obtain reliable performance assessment of the response
surface without additional model evaluations.

Any cross-validation scheme consists in dividing the data sample (i.e.
the experimental design) into two sub-samples that are independently and
identically distributed. A metamodel :math:`\widehat{h}(\underline{x})`
is built from one sub-sample, i.e. the *training set*, and its
performance is assessed by comparing its predictions to the other
subset, i.e. the *test set*. A single split will lead to a validation
estimate. When several splits are conducted, the cross-validation error
estimate is obtained by averaging over the splits.

**K-fold cross-validation error estimate**

The :math:`K`-fold cross-validation technique relies on splitting the
data set :math:`\cX` into :math:`K` mutually exclusive sub-samples
:math:`\cX_1, \dots, \cX_K` of approximately equal size. A sub-sample
:math:`\cX_i`, :math:`i \in \{ 1, \dots, K\}` is set aside, then the
response surface :math:`\widehat{h}^{(-\cX_i)}` is built on the
remaining sub-samples :math:`\cX \setminus \cX_i`. The approximation
error is estimated on the set-aside sample :math:`\cX_i`:

.. math::
   Err^{(i)}  = \dfrac{1}{ |\cX_i|}  \sum\limits_{\underline{x}^{(j)} \in \cX_i} \left[ h(\underline{x}^{(j)}) - \widehat{h}^{(-\cX_i)} {(\underline{x}^{(j)})} \right]^2

in which
:math:`h(\underline{x}^{(j)}) - \widehat{h}^{(-\cX_i)} {(\underline{x}^{(j)})}`
is the *predicted residual* defined as the difference between the
evaluation of :math:`h(\cdot)` and the prediction with
:math:`\widehat{h}^{(-\cX_i)}(\cdot)` at :math:`\underline{x}^{(j)}`
in the sub-sample :math:`\cX_i` whose cardinality is :math:`|\cX_i|`.

The approximation errors :math:`Err^{(i)}` are estimated with each of
the sub-samples :math:`\cX_i`, :math:`i \in \{ 1, \dots, K\}` being
used as the validation set whereas the remaining sub-samples
:math:`\cX \setminus \cX_i` being used for training. Finally the
:math:`K`-fold cross-validation error estimate is obtained as the
average:

.. math::
   Err_{Kfold} = \dfrac{1}{K} \sum\limits_{i=1}^{K} Err^{(i)}

As described above, the :math:`K`-fold error estimate can be obtained
with a single split of the data :math:`\cX` into :math:`K` folds. It
is worth noting that one can repeat the cross-validation multiple
times using different divisions into folds to obtain better Monte
Carlo estimate. This comes obviously with an additional computational
cost.

**Classical leave-one-out error estimate**

The *leave-one-out* (LOO) cross-validation is a special case of
:math:`K`-fold cross-validation where the number of folds :math:`K` is
chosen equal to the cardinality :math:`N` of the experimental design
:math:`\cX`. It is recalled that a :math:`P`-term polynomial
approximation of the model response reads:

.. math:: \widehat{h}(\underline{x}) \, \, = \, \,  \sum_{j=0}^{P-1} \; \widehat{a}_{j} \; \psi_{j}(\underline{x})

where the :math:`\widehat{a}_{j}`\ ’s are estimates of the
coefficients obtained by a specific method, e.g. least squares.

Let us denote by :math:`\widehat{h}^{(-i)}` the approximation that has
been built from the experimental design
:math:`\cX \setminus \{\underline{x}^{(i)}\}` with the :math:`i`-th
observation :math:`\underline{x}^{(i)}` being set aside. The predicted
residual is defined as the difference between the model evaluation at
:math:`\underline{x}^{(i)}` and its prediction based on
:math:`\widehat{h}^{(-i)}`:

.. math::
  :label: 4.3:5

    \Delta^{(i)} \, \, = \, \,  h(\underline{x}^{(i)}) - \widehat{h}^{(-i)}(\underline{x}^{(i)})

By repeating this process for all observations in the experimental
design, one obtains the predicted residuals
:math:`\Delta^{(i)}, i = 1, \dots , N`. Finally, the LOO error is
estimated as follows:

.. math::
  :label: 4.3:6

    Err_{LOO} \, \, = \, \, \frac{1}{N} \sum_{i=1}^{N} {\Delta^{(i)}}^{2}

Due to the linear-in-parameters form of the polynomial chaos
expansion, the quantity :math:`Err_{LOO}` may be computed without
performing further regression calculations when the PC coefficients
have been estimated using the entire experimental design :math:`\cX`.
Indeed, the predicted residuals can be obtained analytically as
follows:

.. math::
  :label: 4.3:7

    \Delta^{(i)} \, = \,
       \frac{h(\underline{x}^{(i)}) - \widehat{h}(\underline{x}^{(i)})}{1 - h_i}

where :math:`h_i` is the :math:`i`-th diagonal term of the matrix
:math:`\underline{\underline{\Psi}} (\underline{\underline{\Psi}}^{\textsf{T}}\underline{\underline{\Psi}})^{-1} \underline{\underline{\Psi}}^{\textsf{T}}`
with :math:`\underline{\underline{\Psi}}` being the information
matrix:

  .. math::
    :label: 4.3:7bis

      \underline{\underline{\Psi}} \, \, = \, \, \left\{ \psi_{j}(\underline{x}^{(i)}) \; , \; i=1,\dots,N \; , \; j = 0,\dots,P-1 \right\}

In practice, one often computes the following normalized LOO error:

  .. math::
    :label: 4.3:8bis

      \varepsilon_{LOO} \, \, \equiv \, \, \frac{Err_{LOO}}{\hat{\Cov{\cY}}}

where :math:`\hat{\Cov{\mathcal{Y}}}` denotes the empirical
covariance of the response sample :math:`\cY`:

  .. math::
    :label: 4.3:4bis

      \hat{\Cov{\mathcal{Y}}} \, \, \equiv \, \, \frac{1}{N-1} \; \sum_{i=1}^{N} \; \left( y^{(i)} \; - \;   \bar{\mathcal{Y}}  \right)^{2} \quad  \quad , \quad \quad   \bar{\mathcal{Y}} \, \, \equiv \, \, \frac{1}{N} \; \sum_{i=1}^{N} \; y^{(i)}

**Corrected leave-one-out error estimate**

A penalized variant of :math:`\varepsilon_{LOO}` may be used in order to
increase its robustness with respect to overfitting, i.e. to penalize a
large number of terms in the PC expansion compared to the size of the
experimental design:

.. math:: \varepsilon_{LOO}^{*} \, \, \equiv \, \, \varepsilon_{LOO} \, T(P,N)

The penalty factor is defined by:

.. math:: T(P,N) \, \, = \, \,   \frac{N}{N-P}  \; \left(1 \; + \; \frac{\mbox{tr} \left( \underline{\underline{C}}_{emp}^{-1}  \right) }{N} \right)

where:

.. math::
  :label: 4.3:10bis

    \underline{\underline{C}}_{emp} \, \, \equiv \, \, \frac{1}{N}\underline{\underline{\Psi}}^{\textsf{T}}\; \underline{\underline{\Psi}} \quad \quad , \quad \quad
    \underline{\underline{\Psi}} \, \, = \, \, \left\{ \psi_{j}(\underline{x}^{(i)}) \, \, , \, \, i=1,\dots,N \, \, , \, \, j=0,\dots,P-1 \right\}


Leave-one-out cross validation is also known as jackknife in statistics.

.. topic:: API:

    - See :class:`~openturns.MetaModelValidation`


.. .. topic:: Examples:
..
..     - See :doc:`/examples/data_analysis/XXXXXXXXXXXXx`


.. topic:: References:

    - [blatman2009]_