I am trying to do text classification using ensemble of transformers. My dataset contains about 64000 tweets. For each tweet i got prediction as follows, So for each tweet each transformer model is giving probabilities whether a tweet a sarcastic or not. As you can see below there are 10 probabilities for each tweet.

Tweet 1 : tensor([0.4219, 0.5781, 0.5237, 0.4763, 0.4977, 0.5023, 0.4618, 0.5382, 0.5324,0.4676])

Tweet 2 : tensor([0.4295, 0.5705, 0.4761, 0.5239, 0.4859, 0.5141, 0.4979, 0.5021, 0.5025,0.4975])

Tweet 3 : tensor([0.4000, 0.6000, 0.4832, 0.5168, 0.4932, 0.5068, 0.5023, 0.4977, 0.4939,0.5061])

.

.

.

Tweet 64000: tensor([0.4213, 0.5787, 0.4904, 0.5096, 0.4870, 0.5130, 0.5084, 0.4916, 0.4900,0.5100])

I need to feed in these probabilities to another neural network or apply logistic regression so that i can get the final prediction for each row. I am not sure how i can achieve these or whether it is actually possible to do. Can you please look into it and provide some insights?

Your answer will be much appreciated.