Added provisoin for label length argument

mlmc/models/abstracts/abstracts.py

@@ -22,7 +22,7 @@ class TextClassificationAbstract(torch.nn.Module):
     """
 
     def __init__(self, classes, target="multi", representation="google/bert_uncased_L-2_H-128_A-2",
-                 activation=None, loss=None, optimizer=torch.optim.Adam, max_len=200,
+                 activation=None, loss=None, optimizer=torch.optim.Adam, max_len=200, label_len=20,
                  optimizer_params=None, device="cpu", finetune=False, threshold="mcut", n_layers=1, **kwargs):
         """
         Abstract initializer of a Text Classification network.
@@ -30,6 +30,7 @@ class TextClassificationAbstract(torch.nn.Module):
             classes: A dictionary of classes and ther corresponding index. This argument is mandatory.
             representation: The string of the input representation. (Supporting the full transformers list, and glove50, glove100, glove200, glove300)
             max_len: The maximum number of tokens for the input.
+            label_len: The maximum number of tokens for labels.
             target: single label oder multilabel mode. defined by keystrings: ("single", "multi").
             Sets some basic options, like loss function, activation and
                     metrics to sensible defaults.
@@ -91,7 +92,8 @@ class TextClassificationAbstract(torch.nn.Module):
             "activation":self.activation, "loss": self.loss,
             "optimizer": self.optimizer, "max_len": self.max_len,
             "optimizer_params": self.optimizer_params, "device": self.device,
-            "finetune": finetune, "threshold": threshold, "n_layers": self.n_layers
+            "finetune": finetune, "threshold": threshold, "n_layers": self.n_layers,
+            "label_len":label_len,
         }
         self._config.update(kwargs)