CODEWITHSUNDEEP
pythonnumpytensorflowneural-networkkeras
Mershel
Mershel

Reputation: 552

Predicting next numbers in sequence Keras - Python

I'm new to python and neural networks. I have a simple network written in Keras that can predict the next number in a linear sequence:

import numpy as np

from keras.models import Sequential
from keras.layers import Dense
from keras.layers import LSTM

data = [[i for i in range(6)]];
data = np.array(data, dtype=int);
target = [[i for i in range(10, 16)]];
target = np.array(target, dtype=int);

model = Sequential();
model.add(Dense(1, input_dim=1))
model.add(Dense(1));
model.compile(loss='mean_absolute_error', optimizer = 'adam', metrics = ['accuracy']);
model.summary();
for i in range (10000):
    dataIterator = 0;
    for targetValue in target:
        model.train_on_batch(data[dataIterator], targetValue)
        dataIterator = dataIterator + 1;

predict = model.predict([28]);
print(predict);

Gives me output:

[[38.0199]]

And that is to be expected. I'm not sure if my code has some stupid errors in it and would appreciate any feedback and explanations. I used Dense because I'm not sure what LSTM exacly does. Another thing is that my model requires input to have 2 dimensions when it's specified: 

input_dim=1

I don't understand why. Next, I would like to create a network that can predict the next numbers in a sequence like [1, 4, 9, 16, 25]. This one doesn't.

Note that this is my first program written in Python and first use of neural networks :). Thanks in advance!

UPDATE 1

According to tip of using scale i came up with something like this:

import numpy as np

from keras.models import Sequential
from pandas import Series
from keras.layers import Dense
from keras.layers import LSTM
from sklearn.preprocessing import StandardScaler

data = [[i for i in range (1,30)]];
data = np.array(data, dtype=int);
target = np.power(data, 2);
target = np.array(target, dtype=int);
target = target.reshape((len(target[0]), 1))
data = data.reshape((len(data[0]), 1))

scale = StandardScaler()

dataTest = [[i for i in range (2,4)]];
dataTest = np.array(dataTest, dtype=int);
dataTest = dataTest.reshape((len(dataTest[0]), 1))

model = Sequential();
model.add(Dense(1, input_dim=1))
model.add(Dense(1));
model.compile(loss='mean_absolute_error', optimizer = 'adam');
model.fit(scale.fit_transform(data), target, batch_size=1, epochs=200, verbose=1)

print(model.predict(scale.transform(dataTest)));

Despite this predictions are way off ideal. For given test data, output: 

[[27.616932]
 [28.265278]]

I'm out of ideas at the moment :(. Not feeling it at all.

Upvotes: 4

Views: 11991

Answers (1)

Mihai Alexandru-Ionut
Mihai Alexandru-Ionut

Reputation: 48367

And that is to be expected. I'm not sure if my code has some stupid errors in it, would appreciate any feedback and explanations. I used Dense because im not sure what LSTM exactly does.

LSTM, which stands for Long-short term memory, are a special kind of RNN, capable of learning long-term dependencies. Also, its use mainly sequential processing over time. For instance, you can use LSTM when you want to predict the Google Stock prices.

Since you have to predict next numbers in sequence like [1, 4, 9, 16, 25] that's means it's a regression learning system model, which belongs to supervised learning. When you're using regression models, there is no accuracy. The accuracy from regression models it's called COD or Coefficient of determination or R squared score.

The metric that you use- metrics=['accuracy'] corresponds to a classification problem. If you want to do regression, remove metrics=['accuracy']. That is, just use

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

Upvotes: 8

Related Questions