Negative Prediction Values

TensorFlow General Discussion
, ,
1

I’m attempting to predict traffic using an LSTM model, but the outcome of my predictions are negative values. Is this normal, or could it indicate an error? Following are my code:

train_predict = model.predict(X_train)
test_predict = model.predict(X_test)

train_predict = scaler.inverse_transform(train_predict)
Y_train = scaler.inverse_transform([Y_train])
test_predict = scaler.inverse_transform(test_predict)
Y_test = scaler.inverse_transform([Y_test])

train_score = np.sqrt(mean_squared_error(Y_train[0], train_predict[:,0]))
print(‘Train RMSE:’, train_score)
test_score = np.sqrt(mean_squared_error(Y_test[0], test_predict[:,0]))
print(‘Test RMSE:’, test_score)

future_time_steps = 5
future_predictions =
last_sequence = X_test[-1]
for _ in range(future_time_steps):
prediction = model.predict(last_sequence.reshape(1, time_steps, 1))
future_predictions.append(prediction[0][0])
last_sequence = np.roll(last_sequence, -1)
last_sequence[-1] = prediction[0][0]
future_predictions = scaler.inverse_transform(np.array(future_predictions).reshape(-1,1))
print(‘Future Predictions:’, future_predictions)

and the result is:
1/1 [==============================] - 0s 24ms/step
1/1 [==============================] - 0s 19ms/step
1/1 [==============================] - 0s 19ms/step
1/1 [==============================] - 0s 18ms/step
1/1 [==============================] - 0s 18ms/step
Future Predictions: [[-1.0864409 ]
[-1.0021602 ]
[-0.9143658 ]
[-0.81085765]
[-0.69729006]]

2

Hi. @Narsis
Maybe you can share the whole code and comment on your dataset? Sharing with us a Colab for instance would be useful.
Thank you.

6

Questions. What is your last layer model.add(Dense(1,activation=‘relu’)) used for? And what if you remove it?

10

Why do you use “relu” activation in the final layer? You might instead try as your final layer model.add(Dense(1)), which defaults to linear activation.