How to prepare the Dataset for Keras LSTM Model?

Question

I want to forecast the future sales for products.

My Questions:

1. How I have to set up the dataset for time series forecasting?
2. What I have to change, to get int values from output?

Model and additional infos

X.shape(68, 108, 124)

y.shape(68, 108, 1)

X
array([[2008,    0,    0, ...,    0,    0,    0],
        [2008,    0,    0, ...,    0,    0,    5],
        [2008,    0,    0, ...,    0,    0,    0],
        ...,
        [2008,    0,    0, ...,    0,    1,    0],
        [2008,    0,    1, ...,    0,    0,    7],
        [2008,    1,    0, ...,    0,    0,    8]],

       [[2009,    0,    0, ...,    0,    0,    0],
        [2009,    0,    0, ...,    0,    0,    0],
        [2009,    0,    0, ...,    0,    1,    10],
        ...,
        [2009,    0,    0, ...,    0,    0,    20],
        [2009,    0,    1, ...,    0,    2,    0],
        [2009,    1,    0, ...,    0,    0,    5]]])

the hole dataset have 68 sequences (each historic year = one sequence), each sequence 108 rows and 124 features. Features: year(int) + countries are one hot encoded (108 different) + 15 product categories (one hot encoded)

output

model.predict(X_train_lstm)

       [[-0.00230993, -0.5376476 , -0.15334567, ...,  0.09862291,
          0.16895522, -0.15376607],
        [ 0.09281429, -0.61930186, -0.30336168, ...,  0.03798292,
          0.19980578, -0.11670589],
        [ 0.10739403, -0.6318171 , -0.32635477, ...,  0.02868856,
          0.20453428, -0.11102566],
        ...,
        [ 0.10971781, -0.63381183, -0.33001956, ...,  0.02720715,
          0.2052879 , -0.11012031],
        [ 0.10971781, -0.63381183, -0.33001956, ...,  0.02720715,
          0.20528793, -0.11012033],
        [ 0.1097179 , -0.63381183, -0.33001956, ...,  0.02720721,
          0.2052879 , -0.1101203 ]],

       ...,

X.shape(68, 108, 124)
y.shape(68, 108, 1)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=seed, shuffle=False)
X_validation, X_test, y_validation, y_test = train_test_split(X_test, y_test, test_size=0.5, random_state=seed, shuffle=False)

model = Sequential()
model.add(LSTM(units=50, input_shape=(108,124), name="LSTM", return_sequences=True))
model.add(Dense(1))
model.compile(optimizer='adam', loss='mean_squared_error', metrics=['accuracy'])
model.fit(X_train, y_train, epochs=10, batch_size=20, verbose=1, validation_data=(X_validation, y_validation))

Answer 1

keras LSTM time series uses vectors of data looking back or forward. you create the dataset with overlapping vectors.

def create_dataset(dataset, look_back=3):
    dataX, dataY = [], []
    for i in range(len(dataset)-look_back-1):
        a = dataset[i:(i+look_back)]
        dataX.append(a)
        dataY.append(dataset[i + look_back])
    return np.array(dataX), np.array(dataY)


COLUMNS=['open']
dataset=df[COLUMNS]
#dataset=np.array(dataset).reshape(-1,1)
scaler = MinMaxScaler(feature_range=(0, 1))
dataset = scaler.fit_transform(np.array(dataset).reshape(-1,1))
#dataset=np.array(pd.DataFrame({"C1":[1,2,3,4,5,6],"C2":[-1,-2,-3,-4,-5,-6],"Y":[7,8,9,10,11,12]})).reshape(-1,1)

train_size = int(len(dataset) * 0.60)
test_size = len(dataset) - train_size
train, test = dataset[0:train_size], dataset[train_size:len(dataset)]
#print(len(train), len(test))

look_back=3
trainX=[]
testX=[]
y_train=[]

trainX, y_train = create_dataset(train, look_back)
testX, y_test = create_dataset(test, look_back)

#print(y_train)

X_train = np.reshape(trainX, (trainX.shape[0], 1, trainX.shape[1]))
X_test = np.reshape(testX, (testX.shape[0], 1, testX.shape[1]))


n_future = 1
features=2
timeSteps=4

model = Sequential()

model.add(Bidirectional(LSTM(units=50, return_sequences=True, 
                             input_shape=(X_train.shape[1], 1))))

model.add(LSTM(units= 50, return_sequences=True))
model.add(Dropout(0.2))
model.add(LSTM(units= 50, return_sequences=True))
model.add(Dropout(0.2))
model.add(LSTM(units= 50))
model.add(Dropout(0.2))
model.add(Dense(units = n_future))

model.compile(optimizer="adam", loss="mean_squared_error", metrics=["acc"])



plot_model(model, to_file='model.png')
img=plt.imread('model.png')
plt.imshow(img)
plt.show()