Merge branch 'develop' into fix_install

pypesto/util.py

@@ -6,7 +6,8 @@
 
 """
 from numbers import Number
-from typing import Any, Callable, Optional, Tuple, Union
+from operator import itemgetter
+from typing import Any, Callable, Optional, Sequence, Tuple, Union
 
 import numpy as np
 from scipy import cluster
@@ -241,7 +242,7 @@ def assign_clusters(vals):
 
 def delete_nan_inf(
     fvals: np.ndarray,
-    x: Optional[np.ndarray] = None,
+    x: Optional[Sequence[Union[np.ndarray, list[float]]]] = None,
     xdim: Optional[int] = 1,
     magnitude_bound: Optional[float] = np.inf,
 ) -> Tuple[np.ndarray, np.ndarray]:
@@ -272,16 +273,17 @@ def delete_nan_inf(
     fvals = np.asarray(fvals)
     finite_fvals = np.isfinite(fvals) & (np.absolute(fvals) < magnitude_bound)
     if x is not None:
-        # if we start out with a list of x, the x corresponding
-        # to finite fvals may be None, so we cannot stack the x before taking
-        # subindexing
-        # If none of the fvals are finite, np.vstack will fail and np.take
-        # will not yield the correct dimension, so we try to construct an
-        # empty np.ndarray with the correct dimension (other functions rely
-        # on x.shape[1] to be of correct dimension)
         if np.isfinite(fvals).any():
-            x = np.vstack(np.take(x, np.where(finite_fvals)[0], axis=0))
+            finite_indices = np.where(finite_fvals)[0]
+            x = np.array(itemgetter(*finite_indices)(x))
+            # itemgetter does not return list for single element
+            if len(x.shape) == 1:
+                x = x.reshape((1, x.shape[0]))
         else:
+            # If none of the fvals are finite we try to construct an
+            # empty np.ndarray with the correct dimension (other functions rely
+            # on x.shape[1] to be of correct dimension)
+            x = np.array(x)
             x = np.empty(
                 (
                     0,

pypesto/visualize/optimization_stats.py

@@ -375,7 +375,7 @@ def stats_lowlevel(
         The plot axes.
     """
     fvals = result.optimize_result.fval
-    values = [res[property_name] for res in result.optimize_result.list]
+    values = [[res[property_name]] for res in result.optimize_result.list]
     values, fvals = delete_nan_inf(fvals, values)
 
     if start_indices is not None:

pypesto/visualize/parameters.py

@@ -115,6 +115,17 @@ def scale_parameters(x):
             start_indices=start_indices,
             plot_inner_parameters=plot_inner_parameters,
         )
+
+        # parse fvals and parameters
+        fvals = np.array(fvals)
+        # remove nan or inf values
+        xs, fvals = delete_nan_inf(
+            fvals=fvals,
+            x=xs,
+            xdim=len(ub) if ub is not None else 1,
+            magnitude_bound=WATERFALL_MAX_VALUE,
+        )
+
         lb, ub, xs = map(scale_parameters, (lb, ub, xs))
 
         # call lowlevel routine
@@ -223,8 +234,8 @@ def parameter_hist(
 
 
 def parameters_lowlevel(
-    xs: Sequence[Union[np.ndarray, List[float]]],
-    fvals: Union[np.ndarray, List[float]],
+    xs: np.ndarray,
+    fvals: np.ndarray,
     lb: Optional[Union[np.ndarray, List[float]]] = None,
     ub: Optional[Union[np.ndarray, List[float]]] = None,
     x_labels: Optional[Iterable[str]] = None,
@@ -241,8 +252,8 @@ def parameters_lowlevel(
     Parameters
     ----------
     xs:
-        Including optimized parameters for each startpoint.
-        Shape: (n_starts, dim).
+        Including optimized parameters for each start that did not result in an infinite fval.
+        Shape: (n_starts_successful, dim).
     fvals:
         Function values. Needed to assign cluster colors.
     lb, ub:
@@ -268,16 +279,6 @@ def parameters_lowlevel(
     ax:
         The plot axes.
     """
-    # parse input
-    xs = np.array(xs)
-    fvals = np.array(fvals)
-    # remove nan or inf values in fvals and xs
-    xs, fvals = delete_nan_inf(
-        fvals=fvals,
-        x=xs,
-        xdim=len(ub) if ub is not None else 1,
-        magnitude_bound=WATERFALL_MAX_VALUE,
-    )
 
     if size is None:
         # 0.5 inch height per parameter

test/visualize/test_visualize.py

@@ -132,10 +132,15 @@ def create_optimization_result_nan_inf():
     result = create_optimization_result()
 
     # append nan and inf
+    # depending on optimizer failed starts's x can be None or vector of np.nan
     optimizer_result = pypesto.OptimizerResult(
         fval=float('nan'), x=np.array([float('nan'), float('nan')]), id='nan'
     )
     result.optimize_result.append(optimize_result=optimizer_result)
+    optimizer_result = pypesto.OptimizerResult(
+        fval=float('nan'), x=None, id='nan_none'
+    )
+    result.optimize_result.append(optimize_result=optimizer_result)
     optimizer_result = pypesto.OptimizerResult(
         fval=-float('inf'),
         x=np.array([-float('inf'), -float('inf')]),
@@ -406,19 +411,21 @@ def test_parameters_with_options(scale_to_interval):
 def test_parameters_lowlevel():
     # create some dummy results
     (lb, ub) = create_bounds()
-    fvals = [0.01, 0.02, 1.01, 2.02, 2.03, 2.04, 3, 4, 4.1, 4.11]
-    xs = [
-        [0.1, 1],
-        [1.2, 3],
-        [2, 4],
-        [1.2, 4.1],
-        [1.1, 3.5],
-        [4.2, 3.5],
-        [1, 4],
-        [6.2, 5],
-        [4.3, 3],
-        [3, 2],
-    ]
+    fvals = np.array([0.01, 0.02, 1.01, 2.02, 2.03, 2.04, 3, 4, 4.1, 4.11])
+    xs = np.array(
+        [
+            [0.1, 1],
+            [1.2, 3],
+            [2, 4],
+            [1.2, 4.1],
+            [1.1, 3.5],
+            [4.2, 3.5],
+            [1, 4],
+            [6.2, 5],
+            [4.3, 3],
+            [3, 2],
+        ]
+    )
 
     # pass lists
     visualize.parameters_lowlevel(xs, fvals, lb=lb, ub=ub)