2
$\begingroup$

Hey everyone,

I am working on a LSTM network in TensorFlow that predicts the values of the price-index of different product-categories in a month, based on those same values of the 12 months before. Unfortunately I ran into some issues while trying to predict. My dataset (data) essentially looks like this:

Product 1 Price-index Product 2 Price-index ... Product 128 Price-index Product 129 Price-index
Month 1 99.1 99.5 ... 100 100.2
Month 2 100.6 101 ... 100.3 100.6
... ... ... ... ... ...
Month 305 150 124 ... 135 136
Month 306 151 126 ... 137 136.2

I did some preprocessing but did not do any rescaling/standardization as the input/outputvalues are all indexes starting from 100 and are therefore on the same scale:

#Split in training and test set
train_size = int(len(data) * 0.9)
test_size = len(data) - train_size
train, test = data.iloc[0:train_size], data.iloc[train_size:len(data)]
print(len(train), len(test))

#Create dataset
def create_dataset(X, time_steps=1):
Xs, ys = [], []
for i in range(len(X) - time_steps):
    v = X.iloc[i:(i + time_steps)].values
    Xs.append(v)
    ys.append(X.iloc[i + time_steps])
return np.array(Xs), np.array(ys)

time_steps = 12
X_train, y_train = create_dataset(train, time_steps)
X_test, y_test = create_dataset(test, time_steps)
print(X_train.shape, y_train.shape)

Which produces the following output:

(263, 12, 129) (263, 129)

My model looks like this:

tf.keras.backend.clear_session

model = tf.keras.Sequential([
  tf.keras.layers.LSTM(50, input_shape=(X_train.shape[1],X_train.shape[2])),
  tf.keras.layers.Dropout(0.2),
  tf.keras.layers.Dense(129)                      
])             

model.compile(loss='mae', optimizer='adam')

history = model.fit(X_train, y_train, epochs=500, batch_size=270, validation_data=(X_test, y_test), verbose=1, shuffle=False)

It gives no error and trains. However, when it is finished training and I try to use predict to get the predicted values for X_train (to plot them against the actual values) I get the following output:

array([[3.1375432, 2.9852755, 3.4141188, ..., 3.3334997, 3.0094063,
    2.5739233],
   [3.1375432, 2.9852755, 3.4141188, ..., 3.3334997, 3.0094063,
    2.5739233],
   [3.1375432, 2.9852755, 3.4141188, ..., 3.3334997, 3.0094063,
    2.5739233],
   ...,
   [3.1375432, 2.9852755, 3.4141188, ..., 3.3334997, 3.0094063,
    2.5739233],
   [3.1375432, 2.9852755, 3.4141188, ..., 3.3334997, 3.0094063,
    2.5739233],
   [3.1375434, 2.9852755, 3.4141188, ..., 3.3334997, 3.0094063,
    2.5739233]], dtype=float32)

Which essentially gives me the exact same output for the different months of prediction. Furthermore, if I try to predict for one month, for example X_train[1] I get the following error from the network:

ValueError: Input 0 of layer sequential_11 is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: (None, 129)

While the shape of X_train = (12,129), the same shape that the network should expect. Is there anyone who can point me in the right direction and show me what I am doing wrong?

Thanks!

$\endgroup$
0
$\begingroup$

it seems like scaling my data helped. I refer to the following thread on GitHub:

https://github.com/keras-team/keras/issues/1727

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.