Merge pull request #56 from mila-iqia/difface_fixes

Difface fixes

crystal_diffusion/models/diffusion_mace.py

@@ -2,7 +2,7 @@
 
 import torch
 from e3nn import o3
-from e3nn.nn import Activation, BatchNorm
+from e3nn.nn import Activation
 from mace.modules import (EquivariantProductBasisBlock, InteractionBlock,
                           LinearNodeEmbeddingBlock, RadialEmbeddingBlock)
 from mace.modules.utils import get_edge_vectors_and_lengths
@@ -146,9 +146,6 @@ def __init__(
             non_linearity = Activation(irreps_in=diffusion_scalar_irreps_out, acts=[gate])
             self.diffusion_scalar_embedding.append(non_linearity)
 
-            normalization = BatchNorm(diffusion_scalar_irreps_out)
-            self.diffusion_scalar_embedding.append(normalization)
-
             linear = o3.Linear(irreps_in=diffusion_scalar_irreps_out,
                                irreps_out=diffusion_scalar_irreps_out,
                                biases=True)

crystal_diffusion/models/scheduler.py

@@ -62,8 +62,7 @@ class CosineAnnealingLRSchedulerParameters(SchedulerParameters):
     eta_min: float = 0.0
 
 
-def load_scheduler_dictionary(hyper_params: SchedulerParameters,
-                              optimizer: optim.Optimizer) -> Dict[AnyStr, Union[optim.lr_scheduler, AnyStr]]:
+def load_scheduler_dictionary(hyper_params: SchedulerParameters, optimizer: optim.Optimizer) -> Dict[AnyStr, Any]:
     """Instantiate the Scheduler.
 
     Args:

crystal_diffusion/utils/sample_trajectory.py

@@ -52,7 +52,7 @@ def read_from_pickle(path_to_pickle: str):
         """Read from pickle."""
         with open(path_to_pickle, 'rb') as fd:
             sample_trajectory = SampleTrajectory()
-            sample_trajectory.data = torch.load(fd)
+            sample_trajectory.data = torch.load(fd, map_location=torch.device('cpu'))
         return sample_trajectory
 
 

examples/mila_cluster/diffusion/config_diffusion_mace.yaml

@@ -30,7 +30,7 @@ model:
     num_interactions: 2
     hidden_irreps: 8x0e + 8x1o
     mlp_irreps: 8x0e
-
+    number_of_mlp_layers: 0
     avg_num_neighbors: 1
     correlation: 3
     gate: silu

experiments/si_diffusion_1x1x1/config_diffusion_mace.yaml

@@ -0,0 +1,85 @@
+# general
+exp_name: difface
+run_name: run1
+max_epoch: 25
+log_every_n_steps: 1
+gradient_clipping: 0.1
+
+# set to null to avoid setting a seed (can speed up GPU computation, but
+# results will not be reproducible)
+seed: 1234
+
+# data
+data:
+  batch_size: 512
+  num_workers: 8
+  max_atom: 8
+
+# architecture
+spatial_dimension: 3
+model:
+  score_network:
+    architecture: diffusion_mace
+    number_of_atoms: 8
+    r_max: 5.0
+    num_bessel: 8
+    num_polynomial_cutoff: 5
+    max_ell: 2
+    interaction_cls: RealAgnosticResidualInteractionBlock
+    interaction_cls_first: RealAgnosticInteractionBlock
+    num_interactions: 2
+    hidden_irreps: 128x0e + 128x1o + 128x2e
+    mlp_irreps: 128x0e
+    number_of_mlp_layers: 0
+    avg_num_neighbors: 1
+    correlation: 3
+    gate: silu
+    radial_MLP: [128, 128, 128]
+    radial_type: bessel
+  noise:
+    total_time_steps: 100
+    sigma_min: 0.001  # default value
+    sigma_max: 0.5  # default value'
+
+# optimizer and scheduler
+optimizer:
+  name: adamw
+  learning_rate: 0.001
+  weight_decay: 1.0e-8
+
+scheduler:
+  name: ReduceLROnPlateau
+  factor: 0.1
+  patience: 20
+
+# early stopping
+early_stopping:
+  metric: validation_epoch_loss
+  mode: min
+  patience: 10
+
+model_checkpoint:
+  monitor: validation_epoch_loss
+  mode: min
+
+# Sampling from the generative model
+diffusion_sampling:
+  noise:
+    total_time_steps: 100
+    sigma_min: 0.001  # default value
+    sigma_max: 0.5  # default value
+  sampling:
+    spatial_dimension: 3
+    number_of_corrector_steps: 1
+    number_of_atoms: 8
+    number_of_samples: 1000
+    sample_every_n_epochs: 5
+    cell_dimensions: [5.43, 5.43, 5.43]
+
+# A callback to check the loss vs. sigma
+loss_monitoring:
+  number_of_bins: 50
+  sample_every_n_epochs: 2
+
+logging:
+  - comet