temp_test

Author: Yufan121

deepmd/entrypoints/train.py

@@ -189,7 +189,7 @@ def _do_work(jdata: Dict[str, Any], run_opt: RunOptions, is_compress: bool = Fal
     assert "training" in jdata
 
     # init the model
-    model = DPTrainer(jdata, run_opt=run_opt, is_compress=is_compress)
+    model = DPTrainer(jdata, run_opt=run_opt, is_compress=is_compress) # call dptrainier 
     rcut = model.model.get_rcut()
     type_map = model.model.get_type_map()
     if len(type_map) == 0:
@@ -324,14 +324,15 @@ def get_data(jdata: Dict[str, Any], rcut, type_map, modifier, multi_task_mode=Fa
         systems=systems,
         batch_size=batch_size,
         test_size=1,  # to satisfy the old api
-        shuffle_test=True,  # to satisfy the old api
+        shuffle_test=False,  # to satisfy the old api   # CHANGED BY YUFAN
         rcut=rcut,
         type_map=type_map,
         optional_type_map=optional_type_map,
         modifier=modifier,
         trn_all_set=True,  # sample from all sets
         sys_probs=sys_probs,
         auto_prob_style=auto_prob,
+        sort_atoms=False,                               # CHANGED BY YUFAN
     )
     data.add_dict(data_requirement)
 

deepmd/loss/ener.py

@@ -171,23 +171,32 @@ def build(self, learning_rate, natoms, model_dict, label_dict, suffix):
             atom_ener_coeff = tf.reshape(atom_ener_coeff, tf.shape(atom_ener))
             energy = tf.reduce_sum(atom_ener_coeff * atom_ener, 1)
         if self.has_e:
-            # Yufan: put extra weight on the last 7 energy components
-            ener_diff = energy - energy_hat # TODO Yufan multiple the weight
-            ener_diff_reshape = tf.reshape(ener_diff, [-1], name='ener_diff_reshape')
-            ener_diff_reshape = tf.concat([ener_diff_reshape[:-7], ener_diff_reshape[-7:] * 10], axis=0) # scale the last 7 force by 10
-            l2_ener_loss = tf.reduce_mean( tf.square(ener_diff), name='l2_'+suffix) # energy loss
+            # # Yufan: put extra weight on the last 7 energy components
+            # ener_diff = energy - energy_hat # TODO Yufan multiple the weight
+            # ener_diff_log = tf.convert_to_tensor(ener_diff, name='ener_diff')
+            # ener_diff_reshape = tf.reshape(ener_diff, [-1], name='ener_diff_reshape')
+            # tf.print("Ener_diff_reshape: ", ener_diff_reshape)
+            # tf.summary.histogram("ener_diff_reshape", ener_diff_reshape)
+            
+            # ener_diff_reshape_shape = tf.strings.format("{}", tf.shape(ener_diff_reshape))
+            # tf.summary.text("ener_diff_reshape_shape", ener_diff_reshape_shape)
+            # ener_diff_log_shape = tf.strings.format("{}", tf.shape(ener_diff_log))
+            # tf.summary.text("ener_diff_log_shape", ener_diff_log_shape)
+            
+            # ener_diff_reshape = tf.concat([ener_diff_reshape[:-7], ener_diff_reshape[-7:] * 10], axis=0) # scale the last 7 force by 10
+            # l2_ener_loss = tf.reduce_mean( tf.square(ener_diff), name='l2_'+suffix) # energy loss
 
-            print("*************************************************************************************************")
-            print("******************** Warning: the last 7 energy components are scaled by 10 ***************")
-            print("*************************************************************************************************")
+            # print("*************************************************************************************************")
+            # print("******************** Warning: the last 7 energy components are scaled by 10 ***************")
+            # print("*************************************************************************************************")
 
-            # print shape of energy and energy_hat
-            print(f"energy shape: {energy.shape}")
-            print(f"energy_hat shape: {energy_hat.shape}")
+            # # print shape of energy and energy_hat
+            # print(f"energy shape: {energy.shape}")
+            # print(f"energy_hat shape: {energy_hat.shape}")
 
-            # l2_ener_loss = tf.reduce_mean(
-            #     tf.square(energy - energy_hat), name="l2_" + suffix
-            # )
+            l2_ener_loss = tf.reduce_mean(
+                tf.square(energy - energy_hat), name="l2_" + suffix
+            )
 
         if self.has_f or self.has_pf or self.relative_f or self.has_gf:
             force_reshape = tf.reshape(force, [-1])
@@ -204,6 +213,16 @@ def build(self, learning_rate, natoms, model_dict, label_dict, suffix):
         if self.has_f:
             # Yufan: put extra weight on the last 7 force components
             diff_f_reshape = tf.reshape(diff_f, [-1], name="diff_f_reshape")
+            diff_f_log = tf.convert_to_tensor(diff_f, name='diff_f')
+            tf.print("Diff_f_reshape: ", diff_f_reshape)
+            tf.summary.histogram("diff_f_reshape", diff_f_reshape)
+            
+            diff_f_reshape_shape = tf.strings.format("{}", tf.shape(diff_f_reshape))
+            tf.summary.text("diff_f_reshape_shape", diff_f_reshape_shape)
+            diff_f_log_shape = tf.strings.format("{}", tf.shape(diff_f_log))
+            tf.summary.text("diff_f_log_shape", diff_f_log_shape)
+            
+            
             diff_f_reshape = tf.concat([diff_f_reshape[:-7*3], diff_f_reshape[-7*3:] * 10], axis=0) # scale the last 7 force by 10
             l2_force_loss = tf.reduce_mean(tf.square(diff_f_reshape), name="l2_force_" + suffix) # force loss