[Time-Series] fix past_observed_mask type (huggingface#22076)

added > 0.5 to `past_observed_mask`

Author: elisim

tests/models/informer/test_modeling_informer.py

@@ -117,7 +117,7 @@ def prepare_informer_inputs_dict(self, config):
 
         past_time_features = floats_tensor([self.batch_size, _past_length, config.num_time_features])
         past_values = floats_tensor([self.batch_size, _past_length])
-        past_observed_mask = floats_tensor([self.batch_size, _past_length])
+        past_observed_mask = floats_tensor([self.batch_size, _past_length]) > 0.5
 
         # decoder inputs
         future_time_features = floats_tensor([self.batch_size, config.prediction_length, config.num_time_features])

tests/models/time_series_transformer/test_modeling_time_series_transformer.py

@@ -114,7 +114,7 @@ def prepare_time_series_transformer_inputs_dict(self, config):
 
         past_time_features = floats_tensor([self.batch_size, _past_length, config.num_time_features])
         past_values = floats_tensor([self.batch_size, _past_length])
-        past_observed_mask = floats_tensor([self.batch_size, _past_length])
+        past_observed_mask = floats_tensor([self.batch_size, _past_length]) > 0.5
 
         # decoder inputs
         future_time_features = floats_tensor([self.batch_size, config.prediction_length, config.num_time_features])