The weekly sales transaction dataset (posted here 
https://web.njit.edu/~usman/courses/cs675_summer20/Sales_Transactions_Dataset_Weekly.csv) 
shows weekly sales of over 800 items across a year. Your task is to use a temporal
convolutional neural network to predict the final week's sales from the previous 
values for each item in the dataset. Report your mean squared error which is defined 
as the mean squared error of your predictions 1/n(sum_i (y'i - yi)**2). The best mean 
squared that we achieve in this dataset is about 17.5 with ridge regression applied 
to an LSTM encoding of the data.

To qualify for full points you should achieve an MSE below 20.

Use the temporal convolutional network in Keras given here 
https://github.com/philipperemy/keras-tcn.

Submit your program that takes as input the dataset 
Sales_Transactions_Dataset_Weekly.csv and outputs the predictions of week 51 for 
each item and the mean squared error.

It's very important that your code does not consider the last column during training.
If it does we will have to assign a grade of 0. If the code is too complicated to
decipher and we cannot tell if you consider the last column we have no choice but
to assign a 0. To avoid such problems make it clear in your code (with comments) that 
you are considering the data only up to week 51 in the training and that week 52 is
clearly specified as test.

Submission directory: /afs/cad/courses/ccs/S20/cs/677/850/<ucid>.
For example if your ucid is abc12 then copy your programs into
/afs/cad/courses/ccs/S20/cs/677/850/abc12.

Your completed program is due before midnight Aug 6th 2020