Hello,
I’m new to pytorch and would like to experiment on a timeseries forecasting problem. My training data consists monthly sales data, a three month moving average, as well as a flag denoting if a sales promotion is happening or not. I am trying to forecast for the next 12 months out.
My training data consists of 37 observations, six lags of all three features with size (37, 6, 3). My target data set is the next 12 months of sales data with size (37, 12).
When I try to convert my arrays into a torch.tensor using the below code, I get the below error:
Xtrain_val_tens = torch.tensor(Xtrain_val, dtype=torch.float64)
Ytrain_val_tens = torch.tensor(Ytrain_val, dtype = torch.float64)
ERROR
TypeError: can't convert np.ndarray of type numpy.object_. The only supported types are: float64, float32, float16, complex64, complex128, int64, int32, int16, int8, uint8, and bool.
I tried the below with no luck:
Xtrain_val_tens = torch.from_numpy(Xtrain_val)
Ytrain_val_tens = torch.from_numpy(Ytrain_val)
If it helps, I start with a dataframe and put said df through the following code to get my arrays:
lags = 6
horizon = 12
###############Make a multi-output time series dataset#####################
series = samp_df[['ord_qty', '3_day_moving_avg', 'promotion_flag']]
#number of lags in the input
Tx = lags
#number of steps in the output
Ty = horizon
#instantiate list to take in the train (X) and test(Y) data set
X = []
Y = []
# run a series from 0 to the length of the dataframe minus the horizon to create a "step through time" data set
for t in range(len(series) - Tx - Ty + 1):
x = series[t:t+Tx]
X.append(x)
y = series['ord_qty'][t+Tx:t+Tx+Ty]
Y.append(y)
# #reshape into numpy arrays for faster processing in ML models
X = np.array(X)
Y = np.array(Y).reshape(-1, Ty)
Xtrain_val, Ytrain_val = X[:-1], Y[:-1]
Xtest_val, Ytest_val = X[-1:], Y[-1:]