diff --git a/egs/wsj/s5/steps/libs/common.py b/egs/wsj/s5/steps/libs/common.py
index 1e8e2ced6ce..503721c23d1 100644
--- a/egs/wsj/s5/steps/libs/common.py
+++ b/egs/wsj/s5/steps/libs/common.py
@@ -18,6 +18,11 @@
 import sys
 import threading
 
+try:
+    import thread as thread_module
+except:
+    import _thread as thread_module
+
 logger = logging.getLogger(__name__)
 logger.addHandler(logging.NullHandler())
 
@@ -230,8 +235,7 @@ def background_command_waiter(command, popen_object, require_zero_status):
             logger.error(str)
             # thread.interrupt_main() sends a KeyboardInterrupt to the main
             # thread, which will generally terminate the program.
-            import thread
-            thread.interrupt_main()
+            thread_module.interrupt_main()
         else:
             logger.warning(str)
 
diff --git a/egs/wsj/s5/steps/libs/nnet3/train/chain_objf/acoustic_model.py b/egs/wsj/s5/steps/libs/nnet3/train/chain_objf/acoustic_model.py
index 229f290e94c..6afb43824fd 100644
--- a/egs/wsj/s5/steps/libs/nnet3/train/chain_objf/acoustic_model.py
+++ b/egs/wsj/s5/steps/libs/nnet3/train/chain_objf/acoustic_model.py
@@ -167,7 +167,7 @@ def train_new_models(dir, iter, srand, num_jobs,
         # work out the 1-based archive index.
         archive_index = (k % num_archives) + 1
         # previous : frame_shift = (k/num_archives) % frame_subsampling_factor
-        frame_shift = ((archive_index + k/num_archives)
+        frame_shift = ((archive_index + k//num_archives)
                        % frame_subsampling_factor)
 
         multitask_egs_opts = common_train_lib.get_multitask_egs_opts(
diff --git a/egs/wsj/s5/steps/libs/nnet3/train/common.py b/egs/wsj/s5/steps/libs/nnet3/train/common.py
index 720164e5436..d052c78b3f8 100644
--- a/egs/wsj/s5/steps/libs/nnet3/train/common.py
+++ b/egs/wsj/s5/steps/libs/nnet3/train/common.py
@@ -288,7 +288,7 @@ def halve_range_str(range_str):
     halved_ranges = []
     for r in ranges:
         # a range may be either e.g. '64', or '128:256'
-        c = [str(max(1, int(x)/2)) for x in r.split(":")]
+        c = [str(max(1, int(x)//2)) for x in r.split(":")]
         halved_ranges.append(":".join(c))
     return ','.join(halved_ranges)
 
@@ -591,7 +591,7 @@ def get_model_combine_iters(num_iters, num_epochs,
         models_to_combine.add(num_iters)
     else:
         subsample_model_factor = 1
-        num_iters_combine = min(max_models_combine, num_iters/2)
+        num_iters_combine = min(max_models_combine, num_iters//2)
         models_to_combine = set(range(num_iters - num_iters_combine + 1,
                                       num_iters + 1))