Forecasting demand n days into the future using Conv1D as a time-series

Question

I have created a Conv1D time series used to forecast weekly demand (my inputs are separated by 7 days). I am now trying to predict the next four weeks of demand. Below is my preprocessing function to achieve the 12 step array:

def df_to_X_y(df, window_size):
    avg = df.mean()
    df_as_np = df.to_numpy()
    X = []
    y = []
    for i in range(len(df_as_np)-window_size):
        row = [[a] for a in df_as_np[i:i+window_size]]
        X.append(row)
        label = df_as_np[i+window_size]
        y.append(label)
    return np.array(X), np.array(y)

I then created a function to iterate thru each dataset and form a prediction for each set:

def conv1D_Time_Series():
    test_data = []
    tndates = []
    ttdates = []
    for i in itemList: 
        train_dates, test_dates = train_test_split(i['WeekDate'], test_size = 0.2, random_state = 25)
        tndates.append(train_dates)
        ttdates.append(test_dates)
   
    for i in itemList:
        training_data, testing_data = train_test_split(i['WeightOrdered'], test_size =0.3, random_state = 25)

        WINDOW_SIZE = 12
        X_train, y_train = df_to_X_y(training_data, WINDOW_SIZE)
        X_test, y_test = df_to_X_y(testing_data, WINDOW_SIZE)
        X_train.shape, y_train.shape, X_test.shape, y_test.shape
        
        modelx = Sequential()
        modelx.add(InputLayer((WINDOW_SIZE,1)))
        modelx.add(Conv1D(64, kernel_size = 2, padding = 'same'))
        modelx.add(Flatten())
        modelx.add(Dense(8, 'relu'))
        modelx.add(Dense(1, 'linear'))

        cpx = ModelCheckpoint('modelx/', save_best_only = True)
        modelx.compile(loss = MeanSquaredError(), optimizer = Adam(learning_rate = 0.0001), metrics = [RootMeanSquaredError()])
        modelx.fit(X_train, y_train, epochs = 500, callbacks = [cpx])
        
        X_FUTURE = 4
        pred = np.array([])
        last = X_test[-12:][0]
        
        for i in range(X_FUTURE):
            curr_prediction = modelx.predict(np.array([last])).flatten()
            print(curr_prediction)
            last = np.concatenate([last[1:], curr_prediction[0]]).flatten()
            pred = np.concatenate([pred, curr_prediction[0]])
        pred = scaler.iverse_transform([pred])[0]
        return pred

The last for-loop block is my attempt at creating this new array of predictions to base further predictions from, yet I am receiving a Value error :

ValueError: all the input arrays must have same number of dimensions, but the array 
at index 0 has 2 dimension(s) and the array at index 1 has 0 dimension(s)

Since my window size is 12, ideally I would like to use the last 12 to predict the first unknown, then 11+prediction to get the 2nd unknown, 10+prediction+prediction to get the 3rd unknown, and so on.

Forecasting demand n days into the future using Conv1D as a time-series

0 Answers0