The FMUFunction class should not exist. Only the FMUPointToFieldFunction class needs to exist. This is because, provided we can access to the whole trajectory of the output depending on the time, then we can always access the last value of the time series. To do this, the user can implement a new FinalValueFMUFunction an OpenTURNSPythonFunction which takes the FMUPointToFieldFunction model as input:
model = otfmi.FMUPointToFieldFunction()
fmuFunction = FinalValueFMUFunction(model)
In the exec() method, we evaluate the underlying function, then extracts the last value of the time series and returns it.
We could, however, make this much easier if a new ot.VertexFunction class is created in OpenTURNS. This function would have the constructor:
function = ot.VertexFunction(pointToFieldFunction, indices)
where pointToFieldFunction is a PointToFieldFunction and indices is:
either a single int if the output dimension of the Field is greater or equal to 1,
or a list of several integers if the output dimension of the Field is equal to 1.
These indices must be in the integer interval $[0, n - 1]$ where $n$ is the number of vertices. This new VertexFunction would return the value of the output Field corresponding to the indices at the vertices corresponding to the given indices.
To make things clearer, I consider the SIR epidemiological model as an example. This is a PointToField function with input dimension 2 corresponding to $\beta$ and $\gamma$ and output the Field corresponding to the time series.
Example 1. Suppose the model only returns the output $S$, i.e. the output Field has dimension equal to 1. Then indices can be a list of arbitrary indices in the integer $[0, n-1]$ where $n$ is the number of time steps. Then the function returns the value of $S$ at the given time points.
Example 2. Suppose the model returns the 3 outputs $S$, $I$ and $R$, i.e. the output Field has dimension equal to 3. Then indices can be a single integer. Then the function returns the value of the vector $(S, I, R)$ at that particular time point. If indices = -1, this corresponds to the last time value.
The
FMUFunctionclass should not exist. Only theFMUPointToFieldFunctionclass needs to exist. This is because, provided we can access to the whole trajectory of the output depending on the time, then we can always access the last value of the time series. To do this, the user can implement a newFinalValueFMUFunctionanOpenTURNSPythonFunctionwhich takes theFMUPointToFieldFunctionmodel as input:In the
exec()method, we evaluate the underlying function, then extracts the last value of the time series and returns it.We could, however, make this much easier if a new
ot.VertexFunctionclass is created in OpenTURNS. This function would have the constructor:where
pointToFieldFunctionis aPointToFieldFunctionandindicesis:intif the output dimension of the Field is greater or equal to 1,These indices must be in the integer interval $[0, n - 1]$ where $n$ is the number of vertices. This new
VertexFunctionwould return the value of the outputFieldcorresponding to the indices at the vertices corresponding to the given indices.To make things clearer, I consider the SIR epidemiological model as an example. This is a
PointToFieldfunction with input dimension 2 corresponding to $\beta$ and $\gamma$ and output theFieldcorresponding to the time series.Fieldhas dimension equal to 1. Thenindicescan be a list of arbitrary indices in the integer $[0, n-1]$ where $n$ is the number of time steps. Then the function returns the value of $S$ at the given time points.Fieldhas dimension equal to 3. Thenindicescan be a single integer. Then the function returns the value of the vector $(S, I, R)$ at that particular time point. Ifindices = -1, this corresponds to the last time value.