Merge pull request #147 from fjosw/fix/non_overlapping_cnfgs

Fix non overlapping configurations

Author: fjosw

pyerrors/input/dobs.py

@@ -397,7 +397,7 @@ def read_pobs(fname, full_output=False, gz=True, separator_insertion=None):
 
 
 # this is based on Mattia Bruno's implementation at https://github.com/mbruno46/pyobs/blob/master/pyobs/IO/xml.py
-def import_dobs_string(content, noempty=False, full_output=False, separator_insertion=True):
+def import_dobs_string(content, full_output=False, separator_insertion=True):
     """Import a list of Obs from a string in the Zeuthen dobs format.
 
     Tags are not written or recovered automatically.
@@ -406,9 +406,6 @@ def import_dobs_string(content, noempty=False, full_output=False, separator_inse
     ----------
     content : str
         XML string containing the data
-    noemtpy : bool
-        If True, ensembles with no contribution to the Obs are not included.
-        If False, ensembles are included as written in the file, possibly with vanishing entries.
     full_output : bool
         If True, a dict containing auxiliary information and the data is returned.
         If False, only the data is returned as list.
@@ -457,7 +454,6 @@ def import_dobs_string(content, noempty=False, full_output=False, separator_inse
     _check(dobs[4].tag == "ne")
     ne = int(dobs[4].text.strip())
     _check(dobs[5].tag == "nc")
-    nc = int(dobs[5].text.strip())
 
     idld = {}
     deltad = {}
@@ -507,7 +503,11 @@ def import_dobs_string(content, noempty=False, full_output=False, separator_inse
 
     for name in names:
         for i in range(len(deltad[name])):
-            deltad[name][i] = np.array(deltad[name][i]) + mean[i]
+            tmp = np.zeros_like(deltad[name][i])
+            for j in range(len(deltad[name][i])):
+                if deltad[name][i][j] != 0.:
+                    tmp[j] = deltad[name][i][j] + mean[i]
+            deltad[name][i] = tmp
 
     res = []
     for i in range(len(mean)):
@@ -516,25 +516,30 @@ def import_dobs_string(content, noempty=False, full_output=False, separator_inse
         obs_names = []
         for name in names:
             h = np.unique(deltad[name][i])
-            if len(h) == 1 and np.all(h == mean[i]) and noempty:
+            if len(h) == 1 and np.all(h == mean[i]):
                 continue
-            deltas.append(deltad[name][i])
-            obs_names.append(name)
-            idl.append(idld[name])
+            repdeltas = []
+            repidl = []
+            for j in range(len(deltad[name][i])):
+                if deltad[name][i][j] != 0.:
+                    repdeltas.append(deltad[name][i][j])
+                    repidl.append(idld[name][j])
+            if len(repdeltas) > 0:
+                obs_names.append(name)
+                deltas.append(repdeltas)
+                idl.append(repidl)
+
         res.append(Obs(deltas, obs_names, idl=idl))
         res[-1]._value = mean[i]
     _check(len(e_names) == ne)
 
     cnames = list(covd.keys())
     for i in range(len(res)):
         new_covobs = {name: Covobs(0, covd[name], name, grad=gradd[name][i]) for name in cnames}
-        if noempty:
-            for name in cnames:
-                if np.all(new_covobs[name].grad == 0):
-                    del new_covobs[name]
-            cnames_loc = list(new_covobs.keys())
-        else:
-            cnames_loc = cnames
+        for name in cnames:
+            if np.all(new_covobs[name].grad == 0):
+                del new_covobs[name]
+        cnames_loc = list(new_covobs.keys())
         for name in cnames_loc:
             res[i].names.append(name)
             res[i].shape[name] = 1
@@ -546,8 +551,6 @@ def import_dobs_string(content, noempty=False, full_output=False, separator_inse
             res[i].tag = symbol[i]
             if res[i].tag == 'None':
                 res[i].tag = None
-    if not noempty:
-        _check(len(res[0].covobs.keys()) == nc)
     if full_output:
         retd = {}
         tool = file_origin.get('tool', None)
@@ -568,7 +571,7 @@ def import_dobs_string(content, noempty=False, full_output=False, separator_inse
         return res
 
 
-def read_dobs(fname, noempty=False, full_output=False, gz=True, separator_insertion=True):
+def read_dobs(fname, full_output=False, gz=True, separator_insertion=True):
     """Import a list of Obs from an xml.gz file in the Zeuthen dobs format.
 
     Tags are not written or recovered automatically.
@@ -577,9 +580,6 @@ def read_dobs(fname, noempty=False, full_output=False, gz=True, separator_insert
     ----------
     fname : str
         Filename of the input file.
-    noemtpy : bool
-        If True, ensembles with no contribution to the Obs are not included.
-        If False, ensembles are included as written in the file.
     full_output : bool
         If True, a dict containing auxiliary information and the data is returned.
         If False, only the data is returned as list.
@@ -615,7 +615,7 @@ def read_dobs(fname, noempty=False, full_output=False, gz=True, separator_insert
         with open(fname, 'r') as fin:
             content = fin.read()
 
-    return import_dobs_string(content, noempty, full_output, separator_insertion=separator_insertion)
+    return import_dobs_string(content, full_output, separator_insertion=separator_insertion)
 
 
 def _dobsdict_to_xmlstring(d):
@@ -782,7 +782,7 @@ def create_dobs_string(obsl, name, spec='dobs v1.0', origin='', symbol=[], who=N
                     o = obsl[oi]
                     if repname in o.idl:
                         if counters[oi] < 0:
-                            num = offsets[oi]
+                            num = 0
                             if num == 0:
                                 data += '0 '
                             else:
@@ -798,7 +798,7 @@ def create_dobs_string(obsl, name, spec='dobs v1.0', origin='', symbol=[], who=N
                             if counters[oi] >= len(o.idl[repname]):
                                 counters[oi] = -1
                         else:
-                            num = offsets[oi]
+                            num = 0
                             if num == 0:
                                 data += '0 '
                             else:

pyerrors/obs.py

@@ -1097,7 +1097,7 @@ def _expand_deltas_for_merge(deltas, idx, shape, new_idx):
     ret = np.zeros(new_idx[-1] - new_idx[0] + 1)
     for i in range(shape):
         ret[idx[i] - new_idx[0]] = deltas[i]
-    return np.array([ret[new_idx[i] - new_idx[0]] for i in range(len(new_idx))])
+    return np.array([ret[new_idx[i] - new_idx[0]] for i in range(len(new_idx))]) * len(new_idx) / len(idx)
 
 
 def derived_observable(func, data, array_mode=False, **kwargs):

tests/json_io_test.py

@@ -339,7 +339,7 @@ def test_dobsio():
 
     dobsio.write_dobs(ol, fname, 'TEST')
 
-    rl = dobsio.read_dobs(fname, noempty=True)
+    rl = dobsio.read_dobs(fname)
     os.remove(fname + '.xml.gz')
     [o.gamma_method() for o in rl]
 

tests/obs_test.py

@@ -566,7 +566,7 @@ def test_intersection_reduce():
     intersection = pe.obs._intersection_idx([o.idl["ens"] for o in [obs1, obs_merge]])
     coll = pe.obs._reduce_deltas(obs_merge.deltas["ens"], obs_merge.idl["ens"], range1)
 
-    assert np.all(coll == obs1.deltas["ens"])
+    assert np.allclose(coll, obs1.deltas["ens"] * (len(obs_merge.idl["ens"]) / len(range1)))
 
 
 def test_irregular_error_propagation():
@@ -878,7 +878,7 @@ def test_correlation_intersection_of_idls():
     cov1 = pe.covariance([obs1, obs2_a])
     corr1 = pe.covariance([obs1, obs2_a], correlation=True)
 
-    obs2_b = obs2_a + pe.Obs([np.random.normal(1.0, 0.1, len(range2))], ["ens"], idl=[range2])
+    obs2_b = (obs2_a + pe.Obs([np.random.normal(1.0, 0.1, len(range2))], ["ens"], idl=[range2])) / 2
     obs2_b.gamma_method()
 
     cov2 = pe.covariance([obs1, obs2_b])
@@ -1038,6 +1038,7 @@ def test_hash():
     assert hash(obs) != hash(o1)
     assert hash(o1) != hash(o2)
 
+
 def test_gm_alias():
     samples = np.random.rand(500)
 
@@ -1049,3 +1050,99 @@ def test_gm_alias():
 
     assert np.isclose(tt1.dvalue, tt2.dvalue)
 
+
+def test_overlapping_missing_cnfgs():
+    length = 200000
+
+    l_samp = np.random.normal(2.87, 0.5, length)
+    s_samp = np.random.normal(7.87, 0.7, length // 2)
+
+    o1 = pe.Obs([l_samp], ["test"])
+    o2 = pe.Obs([s_samp], ["test"], idl=[range(1, length, 2)])
+
+    a2 = pe.Obs([s_samp], ["alt"])
+    t1 = o1 + o2
+    t1.gm(S=0)
+
+    t2 = o1 + a2
+    t2.gm(S=0)
+    assert np.isclose(t1.value, t2.value)
+    assert np.isclose(t1.dvalue, t2.dvalue, rtol=0.01)
+
+
+def test_non_overlapping_missing_cnfgs():
+    length = 100000
+
+    xsamp = np.random.normal(1.0, 1.0, length)
+
+
+    full = pe.Obs([xsamp], ["ensemble"], idl=[range(0, length)])
+    full.gm()
+
+    even = pe.Obs([xsamp[0:length:2]], ["ensemble"], idl=[range(0, length, 2)])
+    odd = pe.Obs([xsamp[1:length:2]], ["ensemble"], idl=[range(1, length, 2)])
+
+    average = (even + odd) / 2
+    average.gm(S=0)
+    assert np.isclose(full.value, average.value)
+    assert np.isclose(full.dvalue, average.dvalue, rtol=0.01)
+
+
+def test_non_overlapping_operations():
+    length = 100000
+
+    samples = np.random.normal(0.93, 0.5, length)
+
+    e = pe.Obs([samples[0:length:2]], ["ensemble"], idl=[range(0, length, 2)])
+    o = pe.Obs([samples[1:length:2]], ["ensemble"], idl=[range(1, length, 2)])
+
+
+    e2 = pe.Obs([samples[0:length:2]], ["even"])
+    o2 = pe.Obs([samples[1:length:2]], ["odd"])
+
+    for func in  [lambda a, b: a + b,
+                  lambda a, b: a - b,
+                  lambda a, b: a * b,
+                  lambda a, b: a / b,
+                  lambda a, b: a ** b]:
+
+        res1 = func(e, o)
+        res1.gm(S=0)
+        res2 = func(e2, o2)
+        res2.gm(S=0)
+
+        print(res1, res2)
+        print((res1.dvalue - res2.dvalue) / res1.dvalue)
+
+        assert np.isclose(res1.value, res2.value)
+        assert np.isclose(res1.dvalue, res2.dvalue, rtol=0.01)
+
+
+def test_non_overlapping_operations_different_lengths():
+    length = 100000
+
+    samples = np.random.normal(0.93, 0.5, length)
+    first = samples[:length // 5]
+    second = samples[length // 5:]
+
+    f1 = pe.Obs([first], ["ensemble"], idl=[range(1, length // 5 + 1)])
+    s1 = pe.Obs([second], ["ensemble"], idl=[range(length // 5, length)])
+
+
+    f2 = pe.Obs([first], ["first"])
+    s2 = pe.Obs([second], ["second"])
+
+    for func in  [lambda a, b: a + b,
+                  lambda a, b: a - b,
+                  lambda a, b: a * b,
+                  lambda a, b: a / b,
+                  lambda a, b: a ** b,
+                  lambda a, b: a ** 2 + b ** 2 / a]:
+
+        res1 = func(f1, f1)
+        res1.gm(S=0)
+        res2 = func(f2, f2)
+        res2.gm(S=0)
+
+        assert np.isclose(res1.value, res2.value)
+        assert np.isclose(res1.dvalue, res2.dvalue, rtol=0.01)