Multivariate time series classification with LSTM (keras)

I have a csv file with ~5000 rows and these columns
“timestamp”,”x1″,”x2″,”x3″,”x4″,”x5″,”x6″,”x7″,”y”
where y is the class [-1,0,1] and x1-7 are the features

I need a python script that creates a new csv file with these columns
“timestamp”,”x1″,”x2″,”x3″,”x4″,”x5″,”x6″,”x7″,”y”,”yhat”
where yhat is the predicted classification

it will use an LSTM model, taking as input 10 past observations
it will use the first 70% of the observations for training and classify the remaining 30% (~1500 rows)

The code must be compatible with the following libraries at the specified versions
scipy: 1.5.2
numpy: 1.19.2
matplotlib: 3.3.2
pandas: 1.1.3
statsmodels: 0.12.0
sklearn: 0.23.2
theano: 1.0.4
Using Theano backend.
keras: 2.2.5