YOU MUST USE THESE FUNCTIONS PLEASE $1)$A function make$\_$tfm$\_$model$($T$)$ which takes a parameter T $($a string with value either $$tcr$$ or

$$antigen$$ and returns a PyTorch object $($untrained$,$ subclass of nn$.$Module$)$ that implements the

transformer model for the corresponding tcr$/$antigen sequences.

$2.$ A function make$\_$predict$\_$model$($M$\_$antigen, M$\_$tcr$)$ which takes two parameters: M$\_$tcr is the

TCR transformer model and M$\_$antigen is the antigen transformer model, and return the full

model that makes the prediction on the TCR$-$antigen interaction.

$3.$ A function pretrain$\_$tfm$\_$model$($M$,$ D$,$ n$\_$epochs$)$ which takes a transformer model M and a

training data set D $($a list of sequences$)$ and pretrain the model M as specified above for

n$\_$epochs. At the end of each epoch, print out the average loss during that epoch.

$4.$ A function train$\_$model$($M$,$ L$\_$antigen, L$\_$tcr$,$ Interaction, n$\_$epochs$)$ which trains the full

prediction model M passed to it for n$\_$epochs using the following data: L$\_$antigen, a list of

antigen sequences, L$\_$tcr$,$ a list of TCR sequences, and Interaction, list of the numbers $(0/1)$

indicating interaction or not. $($Clearly these three lists are of the same length.$)$ At the end of each

epoch, print out the average loss during that epoch.

$5.$ A function load$\_$trained$\_$model$()$ which should download your trained model saved somewhere

online and load it from the downloaded file. The function then returns the model. Do not train the

model from scratch in this function. Same as HW$1/2,$ you should not upload the model file to

moodle. Instead, you should share it in your google drive $($or other online storage which can be

shared by link$).$

$6.$ A function predict$($M$,$ L$\_$antigen, L$\_$tcr$)$ which takes a full prediction model M and L$\_$antigen,

a list of antigen sequences, L$\_$tcr$,$ a list of TCR sequences. $($Again the two lists are of the same

length.$)$ For each corresponding pair of sequences in L$\_$antigen and L$\_$tcr$,$ the function should

use the model to make a prediction on their interaction. Return a list of the prediction results

$(0/1).$

$7.$ In a comment section at the beginning of the $.$py file, report the result from your $3-$fold cross

validations. For each fold$/$run$,$ report the accuracy of the trained model on the left$-$aside test data.

Also report the average of the $3-$folds $($total $4$ numbers$/$measures for a model$).$ Report the

measures for both the model with and the one without pretraining. Show the performance gain by

pretraining and discuss briefly whether it meets your expectation.