unit test number batches for estimate loss raiot

scripts/sh_list_error.sh

@@ -0,0 +1,5 @@
+# find $1 -type f -print0 | xargs -0 grep -li error
+# # B means before, A means after, some erros have long stack exception message so we need at least
+# # 100 lines before the error, the last line usually indicate the root cause of error
+grep -B 100 -wnr "error" --group-separator="=========begin_slurm_error===============" $1 > slurm_errors.txt
+cat slurm_errors.txt

tests/test_number_batches_loss_ratio_estimate.py

@@ -0,0 +1,24 @@
+"""
+loss ratio estimate
+"""
+from tests.utils_test import utils_test_algo
+
+
+def test_loss_ratio_estimate0():
+    """
+    test different number of batches
+    """
+    args = "--te_d=0 --tr_d 2 4 --task=mnistcolor10 --debug --bs=8 \
+        --model=erm --trainer=mldg --nname=conv_bn_pool_2 \
+        --nb4reg_over_task_ratio=0"
+    utils_test_algo(args)
+
+
+def test_loss_ratio_estimate100():
+    """
+    test different number of batches
+    """
+    args = "--te_d=0 --tr_d 2 4 --task=mnistcolor10 --debug --bs=8 \
+        --model=erm --trainer=mldg --nname=conv_bn_pool_2 \
+        --nb4reg_over_task_ratio=100"
+    utils_test_algo(args)