def get_score()

in model/utils/classifier.py [0:0]

• 50 lines of code
• 8 McCabe index (conditional complexity)

    def get_score(self):

        if not self.if_use:
            return 0

        train_sampler = SequentialSampler(self.train_dataset)
        train_dataloader = DataLoader(
            self.train_dataset,
            sampler=train_sampler,
            batch_size=self.batch_size,
        )

        # Train
        self.model.eval()
        X = np.array([])
        y = np.array([])
        for batch in train_dataloader:
            inputs, labels = batch
            inputs = inputs.to(self.device)

            y = np.concatenate([y, labels.cpu().numpy().squeeze()], 0) if y.size else labels.numpy().squeeze()

            with torch.no_grad():
                outputs = self.model(inputs)[0].cpu().numpy().squeeze()
                outputs = np.max(outputs, axis=1)
                X = np.concatenate([X, outputs], 0) if X.size else outputs

        # Define scikit learn model

        # Normalize data
        scaler = StandardScaler()
        scaler.fit(X)
        X = scaler.transform(X)

        clf = svm.LinearSVC(max_iter=10000, dual=False)
        clf.fit(X, y)

        # Compute train_acc
        y_train = clf.predict(X)
        train_acc = accuracy_score(y, y_train)

        # Evaluate

        eval_sampler = SequentialSampler(self.eval_dataset)
        eval_dataloader = DataLoader(
            self.eval_dataset,
            sampler=eval_sampler,
            batch_size=self.batch_size,
        )

        # Eval!
        X = np.array([])
        y = np.array([])
        for batch in eval_dataloader:
            inputs, labels = batch
            inputs = inputs.to(self.device)

            y = np.concatenate([y, labels.cpu().numpy().squeeze()], 0) if y.size else labels.numpy().squeeze()

            with torch.no_grad():
                outputs = self.model(inputs)[0].cpu().numpy().squeeze()
                outputs = np.max(outputs, axis=1)
                X = np.concatenate([X, outputs], 0) if X.size else outputs

        X = scaler.transform(X)
        y_pred = clf.predict(X)

        eval_acc = accuracy_score(y, y_pred)

        self.test_text = None
        self.real_text = None
        torch.cuda.empty_cache()
        return eval_acc