Update example_kuramoto_sivashinsky.py

Correcting code

examples/examples_PINNacle/example_kuramoto_sivashinsky/example_kuramoto_sivashinsky.py

@@ -9,7 +9,6 @@
 import os
 import sys
 import time
-from scipy import interpolate
 
 os.environ['KMP_DUPLICATE_LIB_OK'] = 'TRUE'
 sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..')))
@@ -20,82 +19,28 @@
 from tedeous.callbacks import early_stopping, plot, cache
 from tedeous.optimizers.optimizer import Optimizer
 from tedeous.device import solver_device
+from tedeous.utils import exact_solution_data
+
 
 solver_device('gpu')
 
-KS_datapath = "Kuramoto_Sivashinsky.dat"
+datapath = "Kuramoto_Sivashinsky.dat"
 
 alpha = 100 / 16
 beta = 100 / 16**2
 gamma = 100 / 16**4
 
 
-def func_data(datapath):
-
-    t_transpose = True
-
-    input_dim = 3
-    output_dim = 2
-
-    def trans_time_data_to_dataset(datapath, ref_data):
-        data = ref_data
-        slice = (data.shape[1] - input_dim + 1) // output_dim
-        assert slice * output_dim == data.shape[
-            1] - input_dim + 1, "Data shape is not multiple of pde.output_dim"
-
-        with open(datapath, "r", encoding='utf-8') as f:
-            def extract_time(string):
-                index = string.find("t=")
-                if index == -1:
-                    return None
-                return float(string[index + 2:].split(' ')[0])
-
-            t = None
-            for line in f.readlines():
-                if line.startswith('%') and line.count('@') == slice * output_dim:
-                    t = line.split('@')[1:]
-                    t = list(map(extract_time, t))
-            if t is None or None in t:
-                raise ValueError("Reference Data not in Comsol format or does not contain time info")
-            t = np.array(t[::output_dim])
-
-        t, x0 = np.meshgrid(t, data[:, 0])
-        list_x = [x0.reshape(-1)]
-        for i in range(1, input_dim - 1):
-            list_x.append(np.stack([data[:, i] for _ in range(slice)]).T.reshape(-1))
-        list_x.append(t.reshape(-1))
-        for i in range(output_dim):
-            list_x.append(data[:, input_dim - 1 + i::output_dim].reshape(-1))
-        return np.stack(list_x).T
-
-    scale = 1
-
-    def transform_fn(data):
-        data[:, :input_dim] *= scale
-        return data
-
-    def load_ref_data(datapath, transform_fn=None, t_transpose=False):
-        ref_data = np.loadtxt(datapath, comments="%", encoding='utf-8').astype(np.float32)
-        if t_transpose:
-            ref_data = trans_time_data_to_dataset(datapath, ref_data)
-        if transform_fn is not None:
-            ref_data = transform_fn(ref_data)
-            return ref_data
-
-    load_ref_data(datapath)
-    load_ref_data(datapath, transform_fn, t_transpose)
-
-    return load_ref_data
-
-
 def KS_heterogeneous_experiment(grid_res):
     exp_dict_list = []
 
     x_min, x_max = 0, 2 * np.pi
-    y_min, y_max = 0, 2 * np.pi
     t_max = 1
     grid_res = 20
 
+    pde_dim_in = 2
+    pde_dim_out = 1
+
     domain = Domain()
     domain.variable('x', [x_min, x_max], grid_res)
     domain.variable('t', [0, t_max], grid_res)
@@ -145,7 +90,7 @@ def KS_heterogeneous_experiment(grid_res):
     neurons = 100
 
     net = torch.nn.Sequential(
-        torch.nn.Linear(2, neurons),
+        torch.nn.Linear(pde_dim_in, neurons),
         torch.nn.Tanh(),
         torch.nn.Linear(neurons, neurons),
         torch.nn.Tanh(),
@@ -155,7 +100,7 @@ def KS_heterogeneous_experiment(grid_res):
         torch.nn.Tanh(),
         torch.nn.Linear(neurons, neurons),
         torch.nn.Tanh(),
-        torch.nn.Linear(neurons, 1)
+        torch.nn.Linear(neurons, pde_dim_out)
     )
 
     for m in net.modules():
@@ -169,7 +114,7 @@ def KS_heterogeneous_experiment(grid_res):
 
     model.compile('autograd', lambda_operator=1, lambda_bound=100)
 
-    img_dir = os.path.join(os.path.dirname(__file__), 'diffusion_reaction_2d_gray_scott_img')
+    img_dir = os.path.join(os.path.dirname(__file__), 'kuramoto_sivashinsky_img')
 
     cb_cache = cache.Cache(cache_verbose=True, model_randomize_parameter=1e-6)
 
@@ -180,21 +125,24 @@ def KS_heterogeneous_experiment(grid_res):
                                          randomize_parameter=1e-6,
                                          info_string_every=10)
 
-    cb_plots = plot.Plots(save_every=50,
+    cb_plots = plot.Plots(save_every=500,
                           print_every=None,
                           img_dir=img_dir,
-                          img_dim='3d')  # 3 image dimension options: 3d, 2d, 2d_scatter
+                          img_dim='2d')  # 3 image dimension options: 3d, 2d, 2d_scatter
 
     optimizer = Optimizer('Adam', {'lr': 1e-4})
 
-    model.train(optimizer, 5e6, save_model=True, callbacks=[cb_es, cb_plots, cb_cache])
+    model.train(optimizer, 5e5, save_model=True, callbacks=[cb_es, cb_plots, cb_cache])
 
     end = time.time()
 
     grid = domain.build('NN').to('cuda')
     net = net.to('cuda')
 
-    error_rmse = torch.sqrt(torch.mean((func_data(grid).reshape(-1, 1) - net(grid)) ** 2))
+    exact = exact_solution_data(grid, datapath, pde_dim_in, pde_dim_out).reshape(-1, 1)
+    net_predicted = net(grid)
+
+    error_rmse = torch.sqrt(torch.mean((exact - net_predicted) ** 2))
 
     exp_dict_list.append({'grid_res': grid_res, 'time': end - start, 'RMSE': error_rmse.detach().cpu().numpy(),
                           'type': 'wave_eqn_physical', 'cache': True})
@@ -221,11 +169,3 @@ def KS_heterogeneous_experiment(grid_res):
 # df.boxplot(by='grid_res',column='RMSE',fontsize=42,figsize=(20,10),showfliers=False)
 df.to_csv('examples/benchmarking_data/wave_experiment_physical_10_100_cache={}.csv'.format(str(True)))
 
-
-
-
-
-
-
-
-