importance sampling

espaloma/__init__.py

@@ -3,6 +3,9 @@
 Extensible Surrogate Potential of Ab initio Learned and Optimized by Message-passing Algorithm
 """
 
+import dgl
+import torch
+
 import espaloma.data
 from . import metrics, units
 import espaloma.app

espaloma/mm/functional.py

@@ -24,6 +24,8 @@ def harmonic(x, k, eq, order=[2]):
     if isinstance(order, list):
         order = torch.tensor(order)
 
+    order = order.to(device=x.device)
+
     return k * ((x - eq)).pow(order[:, None, None]).permute(1, 2, 0).sum(
         dim=-1
     )

scripts/force/train_bonded_force.py

@@ -70,6 +70,7 @@ def run(args):
             level='g'
         ),
 
+
         esp.metrics.GraphMetric(
             base_metric=torch.nn.L1Loss(),
             between=['u', 'u_ref'],
@@ -85,12 +86,11 @@ def run(args):
             between=['u', 'u_ref'],
             level='g'
         )
-    ]
 
 
     '''
     metrics_te = [
-        esp.metrics.GraphMetric(
+        esp.metrics.GraphDerivativeMetric(
             base_metric=base_metric,
             between=[param, param + '_ref'],
             level=term
@@ -109,8 +109,8 @@ def run(args):
         metrics_tr=metrics_tr,
         metrics_te=metrics_te,
         n_epochs=args.n_epochs,
-        normalize=esp.data.normalize.PositiveNotNormalize,
-        device=torch.device('cuda:0'),
+        normalize=esp.data.normalize.ESOL100LogNormalNormalize,
+        optimizer=torch.optim.Adam(net.parameters(), 1e-2),
     )
 
     results = exp.run()