Fix CI test based on the requirements of the new merlin loader (#536)

* cast targets to int64 dtype required by torch.scatter_

* update the dataset schema to use the same as the one defined in the model

* update exomrees46 data used in ci with hex feature converted to numerical

ci/test_integration.sh

@@ -35,7 +35,7 @@ pip install -r requirements.txt
 cd t4r_paper_repro
 FEATURE_SCHEMA_PATH=../datasets_configs/ecom_rees46/rees46_schema.pbtxt
 pip install gdown
-gdown https://drive.google.com/uc?id=1payLwuwfa_QG6GvFVg4KT1w7dSkO-jPZ
+gdown https://drive.google.com/uc?id=1NCFZ5ya3zyxPsrmupEoc9UEm4sslAddV
 apt-get update -y
 apt-get install unzip -y
 DATA_PATH=/transformers4rec/examples/t4rec_paper_experiments/t4r_paper_repro/

transformers4rec/torch/losses.py

@@ -31,7 +31,7 @@ def _smooth_one_hot(targets: torch.Tensor, n_classes: int, smoothing: float = 0.
             targets = (
                 torch.empty(size=(targets.size(0), n_classes), device=targets.device)
                 .fill_(smoothing / (n_classes - 1))
-                .scatter_(1, targets.data.unsqueeze(1), 1.0 - smoothing)
+                .scatter_(1, targets.data.unsqueeze(1).to(torch.int64), 1.0 - smoothing)
             )
         return targets
 

transformers4rec/torch/utils/data_utils.py

@@ -225,6 +225,7 @@ def __init__(
             self.drop_last = drop_last
 
             self.set_dataset(buffer_size, engine, reader_kwargs)
+            self.dataset.schema = self.dataset.schema.select_by_name(conts + cats + labels)
 
             if (global_rank is not None) and (self.dataset.npartitions < global_size):
                 logger.warning(

transformers4rec/torch/utils/torch_utils.py

@@ -199,7 +199,7 @@ def create_output_placeholder(scores, ks):
 
 
 def tranform_label_to_onehot(labels, vocab_size):
-    return one_hot_1d(labels.reshape(-1), vocab_size, dtype=torch.float32).detach()
+    return one_hot_1d(labels.reshape(-1).to(torch.int64), vocab_size, dtype=torch.float32).detach()
 
 
 def one_hot_1d(