torch timeseries

# Load dependencies
from sklearn.preprocessing import MinMaxScaler

# Instantiate a scaler
"""
This has to be done outside the function definition so that
we can inverse_transform the prediction set later on.
"""
scaler = MinMaxScaler(feature_range=(-1, 1))

# Extract values from the source .csv file
df = pd.read_csv('../Data/TimeSeriesData/Alcohol_Sales.csv',index_col=0,parse_dates=True)
y = df['S4248SM144NCEN'].values.astype(float)

# Define a test size
test_size = 12

# Create the training set of values
train_set = y[:-test_size]

# DEFINE A FUNCTION:
def create_train_data(seq,ws=12):
    """Takes in a training sequence and window size (ws) of
       default size 12, returns a tensor of (seq/label) tuples"""
    seq_norm = scaler.fit_transform(seq.reshape(-1, 1))    
    seq_norm = torch.FloatTensor(seq_norm).view(-1)

    out = []
    L = len(seq_norm)
    for i in range(L-ws):
        window = seq_norm[i:i+ws]
        label = seq_norm[i+ws:i+ws+1]
        out.append((window,label))
    return out

# Apply the function to train_set
train_data = create_train_data(train_set,12)
len(train_data)  # this should equal 313-12