Use np isclose to check split

code/split_dataset.py

@@ -56,7 +56,7 @@ def hash_withhold(withheld_idxs, length=6):
 def train_tune_test(ds, train_size=.90, tune_size=.1, test_size=0., withhold=None,
                     rseed=8, out_dir=None, overwrite=False):
     """ split data into train, tune, and test sets """
-    if train_size + tune_size + test_size != 1:
+    if not np.isclose(train_size + tune_size + test_size, 1):
         raise ValueError("train_size, tune_size, and test_size must add up to 1. current values are "
                          "tr={}, tu={}, and te={}".format(train_size, tune_size, test_size))