From 8354d765132f6c0ffebfb1e58bb7b15a780d4b3f Mon Sep 17 00:00:00 2001 From: andreapdr Date: Mon, 3 Jul 2023 19:04:26 +0200 Subject: [PATCH] switched from mbert uncased to cased version --- gfun/vgfs/textualTransformerGen.py | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-) diff --git a/gfun/vgfs/textualTransformerGen.py b/gfun/vgfs/textualTransformerGen.py index 93f86b8..6490d08 100644 --- a/gfun/vgfs/textualTransformerGen.py +++ b/gfun/vgfs/textualTransformerGen.py @@ -100,7 +100,7 @@ class TextualTransformerGen(ViewGen, TransformerGen): if "bert" == model_name: return "bert-base-uncased" elif "mbert" == model_name: - return "bert-base-multilingual-uncased" + return "bert-base-multilingual-cased" elif "xlm-roberta" == model_name: return "xlm-roberta-base" elif "mt5" == model_name: @@ -114,12 +114,14 @@ class TextualTransformerGen(ViewGen, TransformerGen): model_name, num_labels=num_labels, output_hidden_states=True ) else: - model_name = "models/vgfs/trained_transformer/mbert-sentiment/checkpoint-8500" # TODO hardcoded to pre-traiend mbert + # model_name = "models/vgfs/trained_transformer/mbert-sentiment/checkpoint-8500" # TODO hardcoded to pre-traiend mbert + model_name = "mbert-rai-multi-2000/checkpoint-1500" # TODO hardcoded to pre-traiend mbert return AutoModelForSequenceClassification.from_pretrained( model_name, num_labels=num_labels, output_hidden_states=True ) def load_tokenizer(self, model_name): + # model_name = "mbert-rai-multi-2000/checkpoint-1500" # TODO hardcoded to pre-traiend mbert return AutoTokenizer.from_pretrained(model_name) def init_model(self, model_name, num_labels): @@ -161,7 +163,7 @@ class TextualTransformerGen(ViewGen, TransformerGen): # split="train", # shuffle=True, # ) - # + # # val_dataloader = self.build_dataloader( # val_lX, # val_lY, @@ -171,9 +173,9 @@ class TextualTransformerGen(ViewGen, TransformerGen): # split="val", # shuffle=False, # ) - # + # # experiment_name = f"{self.model_name.replace('/', '-')}-{self.epochs}-{self.batch_size}-{self.dataset_name}" - # + # # trainer = Trainer( # model=self.model, # optimizer_name="adamW", @@ -202,7 +204,8 @@ class TextualTransformerGen(ViewGen, TransformerGen): # ) if self.probabilistic: - self.feature2posterior_projector.fit(self.transform(lX), lY) + transformed = self.transform(lX) + self.feature2posterior_projector.fit(transformed, lY) self.fitted = True