Merged new changes

FCF.py

@@ -15,5 +15,5 @@
     **blockOps)
 parareal_fcf.plotGraph(N=nBlocks, K=nBlocks,  figSize=(6.4*2, 4.8*2))
 # parareal_fcf.plotGraphForOneBlock(N=nBlocks, K=nBlocks, plotBlock=4, plotIter=2,  figSize=(6.4*2, 4.8*2))
-parareal_fcf.plotSchedule(N=nBlocks, K=nBlocks, nProc=nBlocks, schedulerType="BLOCK-BY-BLOCK")
-
+fig = parareal_fcf.plotSchedule(N=nBlocks, K=nBlocks, nProc=nBlocks, schedulerType="BLOCK-BY-BLOCK")
+fig.show()

README.md

@@ -9,7 +9,8 @@ that uses the [blockops] code for specific application and experiments. Currentl
 https://jupyterhub.mat.tu-harburg.de/blockops/, and here is the current list of WebApp :
 
 - [Accuracy (and stability) of general time-discretization on the Complex Plane](./accuracy)
-- [Automatic Generation of Parareal Task Scheduling using different strategies](./pararealSchedule)
+- [Automatic Generation of Task Scheduling for Parareal](./pararealSchedule)
+- [Automatic Generation of Task Scheduling for Parareal with Overlap (MGRIT-FCF)](./pararealFCFSchedule)
 
 The WebApps are currently under development : current could be improved, and other WebApp could also be added,
 see the [demo WebApp](./demo) for an example.

scripts/07_pararealSpeedup.py

@@ -11,4 +11,5 @@
     "(F - G) u_{n}^k + G * u_{n}^{k+1}",
     propagator="F", predictor="G", F=F, G=G, I=I, name='Parareal')
 
-parareal.plotSchedule(N, K, nProc=N, schedulerType='LOWEST-COST-FIRST')
+fig = parareal.plotSchedule(N, K, nProc=N, schedulerType='LOWEST-COST-FIRST')
+fig.show()

scripts/08_pararealFCFSpeedup.py

@@ -0,0 +1,15 @@
+from blockops import BlockOperator, BlockIteration
+
+G = BlockOperator('G', cost=0.1)
+F = BlockOperator('F', cost=1)
+N = 8
+K = 3
+
+blockOps = dict(F=F, G=G)
+
+parareal_fcf = BlockIteration(
+    update="(F-G)*F*u_{n-1}^k + G*u_{n}^{k+1}",
+    predictor="G", propagator="F", **blockOps)
+fig = parareal_fcf.plotSchedule(
+    N=N, K=K, nProc=N, schedulerType="BLOCK-BY-BLOCK")
+fig.show()

web_apps/pararealFCFSchedule.md

@@ -0,0 +1,3 @@
+## Documentation
+
+... incoming ...

web_apps/pararealFCFSchedule.py

@@ -0,0 +1,139 @@
+from typing import Any
+
+from dynamic_site.app import App, ResponseStages
+from dynamic_site.stage import parameters as par
+import dynamic_site.stage.stages as stages
+
+from blockops import BlockOperator, BlockIteration
+
+# ===================
+# Documentation Stage
+# ===================
+
+s1_docs = stages.DocsStage(
+    "Quick Documentation",
+    r"""
+Consider the block update formula of Parareal with overlap 
+(MGRIT with FCF relaxation) :
+
+$$
+u^{k+1}_{n+1} = F^2 (u^{k}_{n-1}) + G(u^{k+1}_n) - G \circ F(u^{k}_{n-1})
+$$
+
+Define the number of blocks $N$ (time-intervals), the number of iterations $K$ 
+(eventually different for each block), and the cost of $F$
+and $G$.
+Then choose a scheduler type :
+
+- BLOCK-BY-BLOCK : uses $N$ processors, where each processor is dedicated
+to one time block
+- LOWEST-COST-FIRST : uses $N$ processors, and compute first the tasks
+with lower cost
+- OPTIMAL : eventually use more than $N$ processors to minimize the
+overall computation time 
+""",
+)
+
+# ==============
+# Settings Stage
+# ==============
+
+SCHEDULERS = ['LOWEST-COST-FIRST', 'OPTIMAL', 'BLOCK-BY-BLOCK']
+
+pararealSettings = [
+    par.StrictlyPositiveInteger(
+        unique_id="nBlocks", name="$N$",
+        placeholder="Number of Blocks",
+        doc="Strictly positive integer",
+        optional=False,
+    ),
+    par.FloatList(
+        unique_id="nIter", name="$K$",
+        placeholder="Number of Iterations",
+        doc="Number of iteration for each block (1 or N values)",
+        optional=False,
+    ),
+    par.Float(
+        unique_id='costF', name='Computation time for $F$', 
+        placeholder='Floating point number', 
+        doc='Computation time for the fine solver', 
+        optional=False
+    ),
+    par.Float(
+        unique_id='costG', name='Computation time for $G$', 
+        placeholder='Floating point number', 
+        doc='Computation time for the coarse solver', 
+        optional=False
+    ),
+    par.Enumeration(
+        unique_id='schedulerType', name='Scheduler Type', 
+        placeholder='', 
+        doc='Type of scheduler use for Parareal', 
+        optional=False,
+        choices=SCHEDULERS,
+        value='BLOCK-BY-BLOCK'
+    ),
+]
+
+s1_settings = stages.SettingsStage(
+    "pararealSettings", "Parareal Settings", pararealSettings, False
+)
+
+
+# ===========
+# Plots Stage
+# ===========
+
+s1_plot = stages.PlotsStage("Parareal FCF Schedule", None)
+
+# ====================================================
+#                  Accuracy App
+# ====================================================
+class PararealSchedule(App):
+    def __init__(self) -> None:
+        super().__init__(title="Parareal FCF Schedule")
+
+    def compute(self, response_data: dict[str, Any] | None) -> ResponseStages:
+
+        # Create a response, where stages will be added to
+        r = ResponseStages()
+
+        # Initial request
+        r.add_docs_stage(s1_docs)
+        if not response_data:
+            r.add_settings_stage(s1_settings)
+            r.add_plot_stage(s1_plot)
+            return r
+
+        # Convert parameters to correct types
+        pararealSettings = s1_settings.convert_to_types(
+            response_data["pararealSettings"]
+        )
+        # Add the settings with the set values
+        r.add_settings_stage(
+            s1_settings.copy_from_response(response_data["pararealSettings"])
+        )
+
+        G = BlockOperator('G', cost=pararealSettings['costG'])
+        F = BlockOperator('F', cost=pararealSettings['costF'])
+        N = pararealSettings['nBlocks']
+        K = [int(k) for k in pararealSettings['nIter']]
+        K = K[0] if len(K) == 1 else K
+
+        schedulerType = pararealSettings['schedulerType']
+
+        pararealFCF = BlockIteration(
+            "(F - G)*F u_{n-1}^k + G u_{n}^{k+1}",
+            propagator="F", predictor="G", F=F, G=G, name='PararealFCF')
+
+        fig = pararealFCF.plotSchedule(N, K, nProc=N, schedulerType=schedulerType)
+
+        # === Response ===
+
+        plot_stage = s1_plot.copy()
+        plot_stage.plot = fig.to_json()
+        r.add_plot_stage(plot_stage)
+
+        return r
+
+