update develop

README.md

@@ -318,7 +318,7 @@ conda activate neuromancer
 conda install -c defaults intel-openmp -f
 ```
 
-##### MacOS (Apple M1)
+##### MacOS (Apple Silicon)
 
 ``` bash
 conda env create -f osxarm64_env.yml
@@ -352,6 +352,21 @@ From the top level directory of cloned neuromancer
 pip install -e . --no-deps
 ```
 
+### Install Graphviz (optional)
+In order to use the Problem graph plots, we recommend installing Graphviz system-wide. Note that this feature is optional.
+
+#### For Windows:
+Package must be installed manually: [Graphviz website](https://graphviz.org/download/)
+
+#### For Linux (Debian, Ubuntu)
+```bash
+sudo apt install graphviz
+```
+#### For MacOS
+```bash
+brew install graphviz
+```
+
 ### Test NeuroMANCER install
 Run pytest on the [tests folder](https://github.com/pnnl/neuromancer/tree/master/tests). 
 It should take about 2 minutes to run the tests on CPU. 

examples/PDEs/Part_1_PINN_DiffusionEquation.py

@@ -32,6 +32,10 @@
 import numpy as np
 import matplotlib.pyplot as plt
 
+# filter some user warnings from torch broadcast
+import warnings
+warnings.filterwarnings("ignore")
+
 
 def f_real(x, t):
     # exact PDE solution y(x,t)
@@ -61,7 +65,13 @@ def plot3D(X, T, y):
     torch.set_default_dtype(torch.float)    # Set default dtype to float32
     torch.manual_seed(1234)     # PyTorch random number generator
     np.random.seed(1234)        # numpy Random number generators
-    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+    # Device configuration
+    if torch.backends.mps.is_available():
+        device = torch.device('mps')
+    elif torch.cuda.is_available():
+        device = torch.device('cuda')
+    else:
+        device = torch.device('cpu')
 
     """
     ## Generate data of the exact solution
@@ -334,7 +344,7 @@ def plot3D(X, T, y):
                       )
 
     # show the PINN computational graph
-    problem.show()
+    # problem.show()
 
     optimizer = torch.optim.Adam(problem.parameters(), lr=0.001)
     epochs = 5000
@@ -349,6 +359,7 @@ def plot3D(X, T, y):
         dev_metric='train_loss',
         eval_metric="train_loss",
         warmup=epochs,
+        device=device
     )
 
     # Train PINN
@@ -360,7 +371,7 @@ def plot3D(X, T, y):
     Plot the results
     """
     # evaluate trained PINN on test data
-    PINN = problem.nodes[0]
+    PINN = problem.nodes[0].cpu()
     y1 = PINN(test_data.datadict)['y_hat']
 
     # arrange data for plotting
@@ -372,4 +383,3 @@ def plot3D(X, T, y):
     plot3D(X, T, y_real)
     # plot residuals PINN - exact PDE
     plot3D(X, T, y_pinn-y_real)
-

examples/PDEs/Part_2_PINN_BurgersEquation.py

@@ -28,8 +28,15 @@
 import torch.nn as nn
 import numpy as np
 import matplotlib.pyplot as plt
+
 # data imports
 from scipy.io import loadmat
+import os
+os.chdir(os.path.dirname(__file__))
+
+# filter some user warnings from torch broadcast
+import warnings
+warnings.filterwarnings("ignore")
 
 
 def plot3D(X, T, y):
@@ -59,7 +66,14 @@ def plot3D(X, T, y):
     torch.set_default_dtype(torch.float)  # Set default dtype to float32
     torch.manual_seed(1234)  # PyTorch random number generator
     np.random.seed(1234)  # numpy Random number generators
-    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+    # Device configuration
+    if torch.backends.mps.is_available():
+        device = torch.device('mps')
+    elif torch.cuda.is_available():
+        device = torch.device('cuda')
+    else:
+        device = torch.device('cpu')
 
     """
     ## Generate data of the exact solution
@@ -317,7 +331,7 @@ def plot3D(X, T, y):
                       )
 
     # show the PINN computational graph
-    problem.show()
+    # problem.show()
 
     optimizer = torch.optim.AdamW(problem.parameters(), lr=0.003)
     epochs = 8000
@@ -333,6 +347,7 @@ def plot3D(X, T, y):
         dev_metric='train_loss',
         eval_metric="train_loss",
         warmup=epochs,
+        device=device
     )
 
     # Train PINN
@@ -344,7 +359,7 @@ def plot3D(X, T, y):
     Plot the results
     """
     # evaluate trained PINN on test data
-    PINN = problem.nodes[0]
+    PINN = problem.nodes[0].cpu()
     y1 = PINN(test_data.datadict)['y_hat']
 
     # arrange data for plotting

examples/PDEs/Part_3_PINN_BurgersEquation_inverse.py

@@ -30,8 +30,15 @@
 import torch.nn as nn
 import numpy as np
 import matplotlib.pyplot as plt
+
 # data imports
 from scipy.io import loadmat
+import os
+os.chdir(os.path.dirname(__file__))
+
+# filter some user warnings from torch broadcast
+import warnings
+warnings.filterwarnings("ignore")
 
 
 def plot3D(X, T, y):
@@ -61,7 +68,14 @@ def plot3D(X, T, y):
     torch.set_default_dtype(torch.float)  # Set default dtype to float32
     torch.manual_seed(1234)  # PyTorch random number generator
     np.random.seed(1234)  # numpy Random number generators
-    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+
+    # Device configuration
+    if torch.backends.mps.is_available():
+        device = torch.device('mps')
+    elif torch.cuda.is_available():
+        device = torch.device('cuda')
+    else:
+        device = torch.device('cpu')
 
     """
     ## Generate data of the exact solution
@@ -277,7 +291,7 @@ def plot3D(X, T, y):
                       )
 
     # show the PINN computational graph
-    problem.show()
+    # problem.show()
 
     optimizer = torch.optim.AdamW(problem.parameters(), lr=0.001)
     epochs = 10000
@@ -293,6 +307,7 @@ def plot3D(X, T, y):
         dev_metric='train_loss',
         eval_metric="train_loss",
         warmup=epochs,
+        device=device
     )
 
     # Train PINN
@@ -309,7 +324,7 @@ def plot3D(X, T, y):
     Plot the results
     """
     # evaluate trained PINN on test data
-    PINN = problem.nodes[0]
+    PINN = problem.nodes[0].cpu()
     y1 = PINN(test_data.datadict)['y_hat']
 
     # arrange data for plotting