[RTM] New plotting!

README.md

@@ -73,6 +73,8 @@ As a tour guide that uses this binder, you can watch the `bioptim` workshop that
 
     - [OptimalControlProgram](#class-optimalcontrolprogram)
     - [NonLinearProgram](#class-nonlinearprogram)
+    - [VariationalOptimalControlProgram](#class-variationaloptimalcontrolprogram)
+    - [PlottingServer](#class-plottingserver)
 
     </details>
 
@@ -167,6 +169,7 @@ As a tour guide that uses this binder, you can watch the `bioptim` workshop that
     - [Solver](#enum-solver)
     - [ControlType](#enum-controltype)
     - [PlotType](#enum-plottype)
+    - [OnlineOptim](#enum-onlineoptim)
     - [InterpolationType](#enum-interpolationtype)
     - [Shooting](#enum-shooting)
     - [CostType](#enum-costtype)
@@ -787,8 +790,17 @@ The `Solver` class can be used to select the nonlinear solver to solve the ocp:
 
 Note that options can be passed to the solver parameter.
 One can refer to their respective solver's documentation to know which options exist.
-The `show_online_optim` parameter can be set to `True` so the graphs nicely update during the optimization.
-It is expected to slow down the optimization a bit.
+The `show_online_optim` parameter can be set to `True` so the graphs nicely update during the optimization with the default values.
+One can also directly declare `online_optim` as an `OnlineOptim` parameter to customize the behavior of the plotter. 
+Note that `show_online_optim` and `online_optim` are mutually exclusive.
+Please also note that `OnlineOptim.MULTIPROCESS` is not available on Windows and only none of them are available on Macos. 
+To see how to run the server on Windows, please refer to the `getting_started/pendulum.py` example.
+It is expected to slow down the optimization a bit. 
+`show_options` can be also passed as a dict to the plotter to customize the plotter's behavior.
+If `online_optim` is set to `SERVER`, then a server must be started manually by instantiating an `PlottingServer` class (see `ressources/plotting_server.py`).
+The following keys are additional options when using `OnlineOptim.SERVER` and `OnlineOptim.MULTIPROCESS_SERVER`:
+  - `host`: the host to use (default is `localhost`)
+  - `port`: the port to use (default is `5030`)
 
 Finally, one can save and load previously optimized values by using
 ```python
@@ -831,6 +843,12 @@ instead of `x_init` and `x_bounds`.
 the OCP, you can access them with `sol.parameters["qdot_start"]` and `sol.parameters["qdot_end"]` at the end of the
 optimization.
 
+### Class: PlottingServer
+If one wants to use the `OnlineOptim.SERVER` plotter, one can instantiate this class to start a server.
+This is not mandatory as if `as_multiprocess` is set to `True` in the `show_options` dict [default behavior], this server is started automatically.
+The advantage of starting the server manually is that one can plot online graphs on a remote machine.
+An example of such a server is provided in `resources/plotting_server.py`.
+
 ## The model
 
 Bioptim is designed to work with any model, as long as it inherits from the class `bioptim.Model`. Models built with `biorbd` are already compatible with `bioptim`.
@@ -1668,6 +1686,16 @@ INTEGRATED: Plot that links the points within an interval but is discrete betwee
 STEP: Step plot, constant over an interval.
 POINT: Point plot.
 
+### Enum: OnlineOptim
+The type of online plotter to use.
+
+The accepted values are:
+NONE: No online plotter.
+DEFAULT: Use the default online plotter depending on the OS (MULTIPROCESS on Linux, MULTIPROCESS_SERVER on Windows and NONE on MacOS).
+MULTIPROCESS: The online plotter is in a separate process.
+SERVER: The online plotter is in a separate server.
+MULTIPROCESS_SERVER: The online plotter using the server automatically setup on a separate process.
+
 ### Enum: InterpolationType
 Defines wow a time-dependent variable is interpolated.
 It is mainly used for phases time span.

bioptim/__init__.py

@@ -208,6 +208,7 @@
     MagnitudeType,
     MultiCyclicCycleSolutions,
     PhaseDynamics,
+    OnlineOptim,
 )
 from .misc.mapping import BiMappingList, BiMapping, Mapping, NodeMapping, NodeMappingList, SelectionMapping, Dependency
 from .optimization.multi_start import MultiStart
@@ -229,3 +230,5 @@
 from .optimization.stochastic_optimal_control_program import StochasticOptimalControlProgram
 from .optimization.problem_type import SocpType
 from .misc.casadi_expand import lt, le, gt, ge, if_else, if_else_zero
+
+from .gui.online_callback_server import PlottingServer

bioptim/examples/getting_started/pendulum.py

@@ -9,8 +9,6 @@
 appreciate it). Finally, once it finished optimizing, it animates the model using the optimal solution
 """
 
-import platform
-
 from bioptim import (
     OptimalControlProgram,
     DynamicsFcn,
@@ -26,6 +24,7 @@
     BiorbdModel,
     ControlType,
     PhaseDynamics,
+    OnlineOptim,
 )
 
 
@@ -149,8 +148,9 @@ def main():
     # --- Print ocp structure --- #
     ocp.print(to_console=False, to_graph=False)
 
-    # --- Solve the ocp. Please note that online graphics only works with the Linux operating system --- #
-    sol = ocp.solve(Solver.IPOPT(show_online_optim=platform.system() == "Linux"))
+    # --- Solve the ocp --- #
+    # Default is OnlineOptim.MULTIPROCESS on Linux, OnlineOptim.MULTIPROCESS_SERVER on Windows and None on MacOS
+    sol = ocp.solve(Solver.IPOPT(show_online_optim=OnlineOptim.DEFAULT))
 
     # --- Show the results graph --- #
     sol.print_cost()

bioptim/gui/online_callback.py → bioptim/gui/online_callback_abstract.py

@@ -1,12 +1,9 @@
-import multiprocessing as mp
+from abc import ABC, abstractmethod
 
 from casadi import Callback, nlpsol_out, nlpsol_n_out, Sparsity
-from matplotlib import pyplot as plt
 
-from .plot import PlotOcp
 
-
-class OnlineCallback(Callback):
+class OnlineCallbackAbstract(Callback, ABC):
     """
     CasADi interface of Ipopt callbacks
 
@@ -18,12 +15,6 @@ class OnlineCallback(Callback):
         The number of optimization variables
     ng: int
         The number of constraints
-    queue: mp.Queue
-        The multiprocessing queue
-    plotter: ProcessPlotter
-        The callback for plotting for the multiprocessing
-    plot_process: mp.Process
-        The multiprocessing placeholder
 
     Methods
     -------
@@ -37,7 +28,7 @@ class OnlineCallback(Callback):
         Get the name of the output variable
     get_sparsity_in(self, i: int) -> tuple[int]
         Get the sparsity of a specific variable
-    eval(self, arg: list | tuple) -> list[int]
+    eval(self, arg: list | tuple, force: bool = False) -> list[int]
         Send the current data to the plotter
     """
 
@@ -63,20 +54,16 @@ def __init__(self, ocp, opts: dict = None, show_options: dict = None):
         from ..interfaces.ipopt_interface import IpoptInterface
 
         interface = IpoptInterface(ocp)
-        all_g, all_g_bounds = interface.dispatch_bounds()
+        all_g, _ = interface.dispatch_bounds()
         self.ng = all_g.shape[0]
 
-        v = interface.ocp.variables_vector
-
         self.construct("AnimateCallback", opts)
 
-        self.queue = mp.Queue()
-        self.plotter = self.ProcessPlotter(self.ocp)
-        self.plot_process = mp.Process(target=self.plotter, args=(self.queue, show_options), daemon=True)
-        self.plot_process.start()
-
+    @abstractmethod
     def close(self):
-        self.plot_process.kill()
+        """
+        Close the callback
+        """
 
     @staticmethod
     def get_n_in() -> int:
@@ -155,7 +142,8 @@ def get_sparsity_in(self, i: int) -> tuple:
         else:
             return Sparsity(0, 0)
 
-    def eval(self, arg: list | tuple) -> list:
+    @abstractmethod
+    def eval(self, arg: list | tuple, enforce: bool = False) -> list[int]:
         """
         Send the current data to the plotter
 
@@ -164,78 +152,11 @@ def eval(self, arg: list | tuple) -> list:
         arg: list | tuple
             The data to send
 
+        enforce: bool
+            If True, the client will block until the server is ready to receive new data. This is useful at the end of
+            the optimization to make sure the data are plot (and not discarded)
+
         Returns
         -------
         A list of error index
         """
-        send = self.queue.put
-        args_dict = {}
-        for i, s in enumerate(nlpsol_out()):
-            args_dict[s] = arg[i]
-        send(args_dict)
-        return [0]
-
-    class ProcessPlotter(object):
-        """
-        The plotter that interface PlotOcp and the multiprocessing
-
-        Attributes
-        ----------
-        ocp: OptimalControlProgram
-            A reference to the ocp to show
-        pipe: mp.Queue
-            The multiprocessing queue to evaluate
-        plot: PlotOcp
-            The handler on all the figures
-
-        Methods
-        -------
-        callback(self) -> bool
-            The callback to update the graphs
-        """
-
-        def __init__(self, ocp):
-            """
-            Parameters
-            ----------
-            ocp: OptimalControlProgram
-                A reference to the ocp to show
-            """
-
-            self.ocp = ocp
-
-        def __call__(self, pipe: mp.Queue, show_options: dict):
-            """
-            Parameters
-            ----------
-            pipe: mp.Queue
-                The multiprocessing queue to evaluate
-            show_options: dict
-                The option to pass to PlotOcp
-            """
-
-            if show_options is None:
-                show_options = {}
-            self.pipe = pipe
-            self.plot = PlotOcp(self.ocp, **show_options)
-            timer = self.plot.all_figures[0].canvas.new_timer(interval=10)
-            timer.add_callback(self.callback)
-            timer.start()
-            plt.show()
-
-        def callback(self) -> bool:
-            """
-            The callback to update the graphs
-
-            Returns
-            -------
-            True if everything went well
-            """
-
-            while not self.pipe.empty():
-                args = self.pipe.get()
-                self.plot.update_data(args)
-
-            for i, fig in enumerate(self.plot.all_figures):
-                fig.canvas.draw()
-            return True