fix multi target models

megnet/data/graph.py

@@ -1,6 +1,6 @@
 from operator import itemgetter
 import numpy as np
-from megnet.utils.general_utils import expand_1st
+from megnet.utils.general_utils import expand_1st, to_list
 from monty.json import MSONable
 from megnet.data import local_env
 from inspect import signature
@@ -115,7 +115,7 @@ def get_flat_data(self, graphs, targets):
 
         Args:
             graphs: (list of dictionary) list of graph dictionary for each structure
-            targets: (list of float) correpsonding target values for each structure
+            targets: (list of float or list) correpsonding target values for each structure
 
         Returns:
             tuple(node_features, edges_features, global_values, index1, index2, targets)
@@ -125,7 +125,6 @@ def get_flat_data(self, graphs, targets):
         states = []
         index1 = []
         index2 = []
-
         final_targets = []
         for g, t in zip(graphs, targets):
             if isinstance(g, dict):
@@ -134,7 +133,7 @@ def get_flat_data(self, graphs, targets):
                 states.append(g['state'])
                 index1.append(g['index1'])
                 index2.append(g['index2'])
-                final_targets.append([t])
+                final_targets.append(to_list(t))
         return atoms, bonds, states, index1, index2, final_targets
 
     def _get_dummy_convertor(self):

megnet/models.py

@@ -8,7 +8,7 @@
 from megnet.callbacks import ModelCheckpointMAE, ManualStop, ReduceLRUponNan
 from megnet.data.graph import GraphBatchDistanceConvert, GraphBatchGenerator, GaussianDistance
 from megnet.data.crystal import CrystalGraph
-from megnet.utils.preprocessing import StandardScaler
+from megnet.utils.preprocessing import DummyScaler
 import numpy as np
 import os
 from warnings import warn
@@ -40,7 +40,7 @@ class GraphModel:
     def __init__(self,
                  model,
                  graph_convertor,
-                 target_scaler=StandardScaler(mean=0, std=1, is_intensive=True),
+                 target_scaler=DummyScaler(),
                  metadata=None,
                  **kwargs):
         self.model = model
@@ -125,13 +125,11 @@ def train_from_graphs(self,
         callbacks.append(ReduceLRUponNan())
         train_nb_atoms = [len(i['atom']) for i in train_graphs]
         train_targets = [self.target_scaler.transform(i, j) for i, j in zip(train_targets, train_nb_atoms)]
-        train_targets = np.array(train_targets).ravel()
 
         if validation_graphs is not None:
             filepath = pjoin(dirname, 'val_mae_{epoch:05d}_{%s:.6f}.hdf5' % monitor)
             val_nb_atoms = [len(i['atom']) for i in validation_graphs]
             validation_targets = [self.target_scaler.transform(i, j) for i, j in zip(validation_targets, val_nb_atoms)]
-            validation_targets = np.array(validation_targets).ravel()
             val_inputs = self.graph_convertor.get_flat_data(validation_graphs, validation_targets)
 
             val_generator = self._create_generator(*val_inputs,

megnet/tests/test_models.py

@@ -56,6 +56,10 @@ def __getitem__(self, index):
                                 n1=4, n2=4, n3=4, npass=1, ntarget=1,
                                 graph_convertor=CrystalGraph(bond_convertor=GaussianDistance(np.linspace(0, 5, 10), 0.5)),
                                 )
+        cls.model2 = MEGNetModel(10, 2, nblocks=1, lr=1e-2,
+                                 n1=4, n2=4, n3=4, npass=1, ntarget=2,
+                                 graph_convertor=CrystalGraph(bond_convertor=GaussianDistance(np.linspace(0, 5, 10), 0.5)),
+                                 )
 
     def test_train_pred(self):
         s = Structure.from_file(os.path.join(cwd, '../data/tests/cifs/BaTiO3_mp-2998_computed.cif'))
@@ -100,6 +104,13 @@ def test_single_atom_structure(self):
         pred = self.model.predict_structure(s)
         self.assertEqual(len(pred.ravel()), 1)
 
+    def test_two_targets(self):
+        s = Structure(Lattice.cubic(3), ['Si'], [[0, 0, 0]])
+        # initialize the model
+        self.model2.train([s, s], [[0.1, 0.2], [0.1, 0.2]], epochs=1)
+        pred = self.model2.predict_structure(s)
+        self.assertEqual(len(pred.ravel()), 2)
+
     def test_save_and_load(self):
         weights1 = self.model.get_weights()
         with ScratchDir('.'):

megnet/utils/preprocessing.py

@@ -70,3 +70,19 @@ def __str__(self):
 
     def __repr__(self):
         return str(self)
+
+
+class DummyScaler(MSONable):
+    """
+    Dummy scaler does nothing
+    """
+    def transform(self, target, n=1):
+        return target
+
+    def inverse_transform(self, transformed_target, n=1):
+        return transformed_target
+
+    @classmethod
+    def from_training_data(cls, structures, targets, is_intensive=True):
+        return cls()
+