handle nan in tools.golden_sect_DataFrame

pvlib/tests/test_tools.py

@@ -45,3 +45,22 @@ def test__golden_sect_DataFrame_vector():
     v, x = tools._golden_sect_DataFrame(params, lower, upper,
                                         _obj_test_golden_sect)
     assert np.allclose(x, expected, atol=1e-8)
+
+
+def test__golden_sect_DataFrame_nans():
+    # nan in bounds
+    params = {'c': np.array([1., 2., 1.]), 'n': np.array([1., 1., 1.])}
+    lower = np.array([0., 0.001, np.nan])
+    upper = np.array([1.1, 1.2, 1.])
+    expected = np.array([0.5, 0.25, np.nan])
+    v, x = tools._golden_sect_DataFrame(params, lower, upper,
+                                        _obj_test_golden_sect)
+    assert np.allclose(x, expected, atol=1e-8, equal_nan=True)
+    # nan in function values
+    params = {'c': np.array([1., 2., np.nan]), 'n': np.array([1., 1., 1.])}
+    lower = np.array([0., 0.001, 0.])
+    upper = np.array([1.1, 1.2, 1.])
+    expected = np.array([0.5, 0.25, np.nan])
+    v, x = tools._golden_sect_DataFrame(params, lower, upper,
+                                        _obj_test_golden_sect)
+    assert np.allclose(x, expected, atol=1e-8, equal_nan=True)

pvlib/tools.py

@@ -286,14 +286,17 @@ def _golden_sect_DataFrame(params, lower, upper, func, atol=1e-8):
 
     Parameters
     ----------
-    params : dict or Dataframe
-        Parameters to be passed to `func`.
+    params : dict of numeric
+        Parameters to be passed to `func`. Each entry must be of the same
+        length.
 
     lower: numeric
-        Lower bound for the optimization
+        Lower bound for the optimization. Must be the same length as each
+        entry of params.
 
     upper: numeric
-        Upper bound for the optimization
+        Upper bound for the optimization. Must be the same length as each
+        entry of params.
 
     func: function
         Function to be optimized. Must be in the form
@@ -312,6 +315,7 @@ def _golden_sect_DataFrame(params, lower, upper, func, atol=1e-8):
     Notes
     -----
     This function will find the points where the function is maximized.
+    Returns nan where lower or upper is nan, or where func evaluates to nan.
 
     See also
     --------
@@ -326,10 +330,10 @@ def _golden_sect_DataFrame(params, lower, upper, func, atol=1e-8):
 
     converged = False
     iterations = 0
-    iterlimit = 1 + np.max(
+    iterlimit = 1 + np.nanmax(
         np.trunc(np.log(atol / (df['VH'] - df['VL'])) / np.log(phim1)))
 
-    while not converged and (iterations < iterlimit):
+    while not converged and (iterations <= iterlimit):
 
         phi = phim1 * (df['VH'] - df['VL'])
         df['V1'] = df['VL'] + phi
@@ -345,15 +349,18 @@ def _golden_sect_DataFrame(params, lower, upper, func, atol=1e-8):
         err = abs(df['V2'] - df['V1'])
 
         # works with single value because err is np.float64
-        converged = (err < atol).all()
+        converged = (err[~np.isnan(err)] < atol).all()
         # err will be less than atol before iterations hit the limit
         # but just to be safe
         iterations += 1
 
     if iterations > iterlimit:
-        raise Exception("iterations exceeded maximum")  # pragma: no cover
+        raise Exception("Iterations exceeded maximum. Check that func",
+                        " is not NaN in (lower, upper)")
 
-    return func(df, 'V1'), df['V1']
+    func_result = func(df, 'V1')
+    x = np.where(np.isnan(func_result), np.nan, df['V1'])
+    return func_result, x
 
 
 def _get_sample_intervals(times, win_length):