Merge pull request #157 from ICB-DCM/develop

Release 0.0.6

.gitignore

@@ -100,3 +100,4 @@ venv/*
 tmp/*
 amici_models/*
 *.txt
+test/doc/example/tmp/benchmark-models/

.travis.yml

@@ -31,7 +31,8 @@ install:
 # run tests
 script:
   - python3 -m flake8 --exclude=build,doc,example,tmp,amici_models
-  - xvfb-run -a python3 -m pytest --cov=pypesto ./test
+  - travis_wait 20 python3 -m pytest --cov=pypesto ./test/test_*
+  - travis_wait 20 xvfb-run -a python3 -m pytest --cov=pypesto --cov-append ./test/visualize/test_*
   - coverage xml
   - test/run_notebook.sh doc/example/
 
@@ -49,4 +50,3 @@ deploy:
 # cache dependencies
 
 cache: pip
-

doc/releasenotes.rst

@@ -6,6 +6,12 @@ Release notes
 ..........
 
 
+0.0.6 (2019-03-13)
+------------------
+
+* Several minor error fixes, in particular on tests and steady state.
+
+
 0.0.5 (2019-03-11)
 ------------------
 

pypesto/objective/aggregated.py

@@ -1,4 +1,5 @@
 import numpy as np
+from copy import deepcopy
 
 from .objective import Objective
 
@@ -90,6 +91,14 @@ def __init__(self, objectives, x_names=None, options=None):
 
         super().__init__(**init_kwargs)
 
+    def __deepcopy__(self, memodict=None):
+        other = AggregatedObjective(
+            objectives=[deepcopy(objective) for objective in self.objectives],
+            x_names=deepcopy(self.x_names),
+            options=deepcopy(self.options),
+        )
+        return other
+
     def aggregate_fun_sensi_orders(self, x, sensi_orders):
         rvals = [
             objective.fun(x, sensi_orders)
@@ -189,6 +198,15 @@ def aggregate_hess(self, x):
     def aggregate_hessp(self, x):
         return sum(objective.hessp(x) for objective in self.objectives)
 
+    def reset_steadystate_guesses(self):
+        """
+        Propagates reset_steadystate_guesses() to child objectives if available
+        (currently only applies for amici_objective)
+        """
+        for objective in self.objectives:
+            if hasattr(objective, 'reset_steadystate_guesses'):
+                objective.reset_steadystate_guesses()
+
 
 def _check_boolean_value_consistent(objectives, attr):
     values = set(

pypesto/objective/amici_objective.py

@@ -67,11 +67,14 @@ def __init__(self,
 
         guess_steadystate: bool, optional (default = True)
             Whether to guess steadystates based on previous steadystates and
-            respective derivatives.
+            respective derivatives. This option may lead to unexpected
+            results for models with conservation laws and should accordingly
+            be deactivated for those models.
 
         n_threads: int, optional (default = 1)
             Number of threads that are used for parallelization over
-            experimental conditions.
+            experimental conditions. If amici was not installed with openMP
+            support this option will have no effect.
 
         options: pypesto.ObjectiveOptions, optional
             Further options.
@@ -152,11 +155,15 @@ def __init__(self,
         # preallocate guesses, construct a dict for every edata for which we
         # need to do preequilibration
         if self.guess_steadystate:
+            if self.amici_model.ncl() > 0:
+                raise ValueError('Steadystate prediciton is not supported for'
+                                 'models with conservation laws!')
+
             if self.amici_model.getSteadyStateSensitivityMode() == \
                     amici.SteadyStateSensitivityMode_simulationFSA:
                 raise ValueError('Steadystate guesses cannot be enabled when'
                                  ' `simulationFSA` as '
-                                 'SteadyStateSensitivityMode')
+                                 'SteadyStateSensitivityMode!')
             self.steadystate_guesses = {
                 'fval': np.inf,
                 'data': {
@@ -217,7 +224,8 @@ def __deepcopy__(self, memodict=None):
         model = amici.ModelPtr(self.amici_model.clone())
         solver = amici.SolverPtr(self.amici_solver.clone())
         edatas = [amici.ExpData(data) for data in self.edatas]
-        other = AmiciObjective(model, solver, edatas)
+        other = AmiciObjective(model, solver, edatas,
+                               guess_steadystate=self.guess_steadystate)
         for attr in self.__dict__:
             if attr not in ['amici_solver', 'amici_model', 'edatas']:
                 other.__dict__[attr] = copy.deepcopy(self.__dict__[attr])
@@ -245,10 +253,7 @@ def _call_amici(
         if sensi_order > self.max_sensi_order:
             raise Exception("Sensitivity order not allowed.")
 
-        # prepare result objects
-
-        rdatas = []
-
+        # prepare outputs
         nllh = 0.0
         snllh = np.zeros(self.dim)
         s2nllh = np.zeros([self.dim, self.dim])
@@ -280,7 +285,7 @@ def _call_amici(
             self.amici_model,
             self.amici_solver,
             self.edatas,
-            num_threads=self.n_threads
+            num_threads=min(self.n_threads, len(self.edatas)),
         )
 
         for data_ix, rdata in enumerate(rdatas):
@@ -324,7 +329,7 @@ def _call_amici(
                         if sres.size else opt_sres
 
         # check whether we should update data for preequilibration guesses
-        if 'fval' in self.steadystate_guesses and \
+        if self.guess_steadystate and \
                 nllh <= self.steadystate_guesses['fval']:
             self.steadystate_guesses['fval'] = nllh
             for data_ix, rdata in enumerate(rdatas):
@@ -429,9 +434,12 @@ def apply_steadystate_guess(self, condition_ix, x):
             if guess_data['x_ss'] is not None:
                 x_ss_guess = guess_data['x_ss']
             if guess_data['sx_ss'] is not None:
-                x_ss_guess += guess_data['sx_ss'].transpose().dot(
+                linear_update = guess_data['sx_ss'].transpose().dot(
                     (x_sim - guess_data['x'])
                 )
+                # limit linear updates to max 20 % elementwise change
+                if (x_ss_guess/linear_update).max() < 0.2:
+                    x_ss_guess += linear_update
 
         self.edatas[condition_ix].x0 = tuple(x_ss_guess)
 
@@ -685,13 +693,15 @@ def sim_sres_to_opt_sres(par_opt_ids,
     """
     opt_sres = np.zeros((sim_sres.shape[0], len(par_opt_ids)))
 
-    for par_sim_idx, par_opt_id in enumerate(mapping_par_opt_to_par_sim):
+    par_sim_idx = 0
+    for par_opt_id in mapping_par_opt_to_par_sim:
         if not isinstance(par_opt_id, str):
             # this was a numeric override for which we ignore the hessian
             continue
 
         par_opt_idx = par_opt_ids.index(par_opt_id)
         opt_sres[:, par_opt_idx] += \
             coefficient * sim_sres[:, par_sim_idx]
+        par_sim_idx += 1
 
     return opt_sres

pypesto/objective/history.py

@@ -181,7 +181,7 @@ def _update_trace(self, x, sensi_orders, mode, result):
                 or 0 not in sensi_orders \
                 else res_to_chi2(res_result)
             schi2 = None if not self.options.trace_record_schi2 \
-                or 0 not in sensi_orders or 1 not in sensi_orders \
+                or 1 not in sensi_orders \
                 else sres_to_schi2(res_result, sres_result)
 
         # check whether to append to trace

pypesto/objective/util.py

@@ -23,4 +23,4 @@ def sres_to_schi2(res, sres):
     """
     if res is None or sres is None:
         return None
-    return res * sres
+    return res.dot(sres)

pypesto/optimize/optimizer.py

@@ -104,6 +104,8 @@ def objective_decorator(minimize):
 
     def wrapped_minimize(self, problem, x0, index):
         problem.objective.reset_history(index=index)
+        if hasattr(problem.objective, 'reset_steadystate_guesses'):
+            problem.objective.reset_steadystate_guesses()
         result = minimize(self, problem, x0, index)
         problem.objective.finalize_history()
         result = fill_result_from_objective_history(

pypesto/startpoint/util.py

@@ -37,7 +37,7 @@ def assign_startpoints(n_starts, startpoint_method, problem, options):
     n_required_points = n_starts - n_guessed_points
 
     if n_required_points <= 0:
-        return x_guesses[n_starts, :]
+        return x_guesses[:n_starts, :]
 
     # apply startpoint method
     x_sampled = startpoint_method(

pypesto/version.py

@@ -1 +1 @@
-__version__ = "0.0.5"
+__version__ = "0.0.6"

pypesto/visualize/reference_points.py

@@ -174,7 +174,7 @@ def create_references(references=None, x=None, fval=None, color=None,
 
     # parse input (x and fval)
     if (x is not None) and (fval is not None):
-        ref.append(ReferencePoint(x=x, fval=fval, color=color))
+        ref.append(ReferencePoint(x=x, fval=fval, color=color, legend=legend))
 
     # assign colors for reference points which have no user-specified colors
     return assign_colors(ref)

test/test_engine.py

@@ -10,8 +10,7 @@ class EngineTest(unittest.TestCase):
 
     def test_basic(self):
         for engine in [pypesto.SingleCoreEngine(),
-                       pypesto.MultiProcessEngine(),
-                       pypesto.MultiProcessEngine(5)]:
+                       pypesto.MultiProcessEngine(n_procs=2)]:
             self._test_basic(engine)
 
     def _test_basic(self, engine):