Optional param input in function definition (PaddlePaddle#151)

* remove print

* add padding in create input from loss

* fix nproc typo

* update name param to params

* update param in algo

paddlescience/algorithm/algorithm_pinns.py

@@ -511,7 +511,7 @@ def feed_data_user_next(self, labels, labels_attr, data):
             # print("idx user next: ", idx)
         return labels
 
-    def compute_forward(self, *inputs, params=None):
+    def compute_forward(self, params, *inputs):
 
         outs = list()
 
@@ -521,14 +521,8 @@ def compute_forward(self, *inputs, params=None):
 
         return outs
 
-    def compute(self,
-                *inputs_labels,
-                ninputs,
-                inputs_attr,
-                nlabels,
-                labels_attr,
-                pde,
-                params=None):
+    def compute(self, params, *inputs_labels, ninputs, inputs_attr, nlabels,
+                labels_attr, pde):
 
         outs = list()
 
@@ -561,7 +555,7 @@ def compute(self,
                 labels,
                 labels_attr["interior"],
                 bs=-1,
-                params=None)  # TODO: bs is not used
+                params=params)  # TODO: bs is not used
             loss_eq += loss_i
             outs.append(out_i)
             n += 1
@@ -581,7 +575,7 @@ def compute(self,
                 labels,
                 labels_attr,
                 bs=-1,
-                params=None)  # TODO: bs is not used
+                params=params)  # TODO: bs is not used
             loss_bc += loss_b
             outs.append(out_b)
             n += 1
@@ -597,7 +591,7 @@ def compute(self,
                 labels,
                 labels_attr,
                 bs=-1,
-                params=None)
+                params=params)
             loss_ic += loss_it
             outs.append(out_it)
             n += 1
@@ -617,7 +611,7 @@ def compute(self,
                 labels,
                 labels_attr["user"],
                 bs=-1,
-                params=None)
+                params=params)
             loss_eq += loss_id
 
             # data loss
@@ -629,7 +623,7 @@ def compute(self,
                 labels,
                 labels_attr["user"],
                 bs=-1,
-                params=None)  # TODO: bs is not used
+                params=params)  # TODO: bs is not used
             loss_data += loss_d
             outs.append(out_id)
 

paddlescience/solver/solver.py

@@ -18,7 +18,6 @@
 from paddle.distributed.auto_parallel.engine import Engine
 from paddle.incubate.optimizer.functional.lbfgs import minimize_lbfgs
 from paddle.incubate.optimizer.functional.bfgs import minimize_bfgs
-paddle.disable_static()
 from . import utils
 from .. import config
 from visualdl import LogWriter
@@ -56,6 +55,7 @@ def forward(self, *inputs_labels):
             input.stop_gradient = False
 
         self.loss, self.outs, self.loss_details = self.algo.compute(
+            None,
             *inputs_labels,
             ninputs=self.ninputs,
             inputs_attr=self.inputs_attr,
@@ -210,6 +210,7 @@ def __solve_dynamic(self, num_epoch, bs, checkpoint_freq, checkpoint_path):
                 # TODO: error out num_epoch==0
 
                 loss, outs, loss_details = self.algo.compute(
+                    None,
                     *inputs_labels,
                     ninputs=ninputs,
                     inputs_attr=inputs_attr,
@@ -266,6 +267,7 @@ def __solve_dynamic(self, num_epoch, bs, checkpoint_freq, checkpoint_path):
             def _f(x):
                 self.algo.net.reconstruct(x)
                 loss, self.outs, self.loss_details = self.algo.compute(
+                    None,
                     *inputs_labels,
                     ninputs=ninputs,
                     inputs_attr=inputs_attr,
@@ -347,7 +349,7 @@ def __predict_dynamic(self):
             inputs[i] = paddle.to_tensor(
                 inputs[i], dtype=self._dtype, stop_gradient=False)
 
-        outs = self.algo.compute_forward(*inputs)
+        outs = self.algo.compute_forward(None, *inputs)
 
         for i in range(len(outs)):
             outs[i] = outs[i].numpy()
@@ -413,6 +415,7 @@ def __init_static(self):
                     inputs_labels.append(label)
 
                 self.loss, self.outs, self.loss_details = self.algo.compute(
+                    None,
                     *inputs_labels,
                     ninputs=ninputs,
                     inputs_attr=self.inputs_attr,
@@ -543,7 +546,7 @@ def __predict_static(self):
                     input.stop_gradient = False
                     ins.append(input)
 
-                self.outs_predict = self.algo.compute_forward(*ins)
+                self.outs_predict = self.algo.compute_forward(None, *ins)
 
         # startup program
         self.exe.run(self.startup_program)
@@ -652,7 +655,7 @@ def __solve_static_auto_dist(self, num_epoch, bs, checkpoint_freq):
                     input.stop_gradient = False
                     ins.append(input)
 
-                self.outs_predict = self.algo.compute_forward(*ins)
+                self.outs_predict = self.algo.compute_forward(None, *ins)
 
         # feeds inputs
         feeds = dict()