.. _Metrics Discrete Dynamic:
Metrics for the Discrete Dynamic model with GUI
====================================================
The **Metrics** tab calculates a set of metrics on the provided dataset.
Metrics, provided for **Discrete Dynamic** are:
.. math:: \frac{ \|prediction\_ normalized - reference\_ normalized\|_{fro}}{\|reference\_ normalized\|_{fro}}
.. math:: \frac{ \|prediction - reference\|_{fro}}{\|reference\|_{fro}}
.. math:: \frac{ \|prediction - reference\|_{fro}}{\|(reference - mean(reference))\|_{fro}}
.. math:: \frac{max(|prediction - reference|)}{max(|reference|)}
.. math:: \frac{max(|prediction - reference|)}{max(|reference|) - min(|reference|)}
* Switch to the **Metrics** tab
* To calculate metrics, click on the dataset in the **Evaluation files** section. Use **Aditional +** to add the datasets. For metrics calculation both **input file** and **output file** containing the data at the same timepoints have to be provided (see :std:ref:`Data preparation for NeurEco Discrete Dynamic GUI`).
* Choose the initialization option:
* Click **Default Initialization** for metrics with the prediction evaluated using the default initialization: the number of time steps used for initialization is equal to the number deduced during the **Build** of the model (**init time steps** provided in the checkpoint slider).
The **Default Initialization** is a special case of the explicit initialization.
* Click **Steady State Initialization** (to use only when the explicit initialization is not available): the beginning of the trajectory is computed from the steady state deduced from the model; the data in **output file** are not used for evaluation, but only as a **reference** for computation of the metrics.
* To provide an explicit initialization:
* Click on the pair input/output, a window for **Time steps for initialization** appears.
Here, the initialization interval always starts from the beginning of the trajectory.
Choose the number of time steps to use for initialization **Time steps for initialization**. The value proposed by default is equal to **init time steps** from the **Default Initialization**
The interval is taken from **output file** to initialize the evaluation of the prediction (see :std:ref:`Evaluate NeurEco Discrete Dynamic model with GUI`).
* Click on **Compute metrics**
.. figure:: ../../../images/GUIMetricsDynamicInit.png
:width: 400
:alt: GUIMetricsDynamicInit
:align: center
* The results are displayed, and the **Metrics** tab provides also a **Plot reference vs. prediction** for the selected dataset.
An example of a result looks as follows:
.. figure:: ../../../images/GUIMetricsDiscreteDynamic.png
:width: 700
:alt: GUIMetricsDiscreteDynamic
:align: center
GUI operations: metrics evaluation for **Discrete Dynamic**, test case :std:ref:`Nonlinear oscillator test case`
.. Note::
| By default, the evaluation of metrics is performed with the last model available in the checkpoint.
| Use the checkpoint slider in the bottom to choose any other available model and get its metrics.