Deep Learning Model to predict sum of Sequences based on flag value

Question

I am trying to Predict Sum of the Sequence based on flag but my model is not able to converge.

for each time stamp, include the first element in sum if second number is 1 in Sequence.

Example

[[50,1],[40,0]]
[[150,1],[4,1]]
[[60,0],[40,1]]
[[760,0],[400,0]]

Output Should be:

Which model should I use? I am trying lstm using tflearn but result doesn't seems to be good.

Treat these as four features and try LinearRegression. Add interaction features. It will converge. Then remove low importance features. Only AB and CD should come out as important features. Coeff. should come around 1. — 10xAI, Jul 28, 2021 at 12:54
Actually this is a sample task... In original problem I have 10 TimeStep with size 24 at each step. However output will depend only on 2 feature at each timestamp. — Anup Patel, Jul 28, 2021 at 13:52

Nicolas Martin · Accepted Answer · 2021-07-27 15:26:36Z

0

Why don't you use mathematical operations to solve your problem?

[[50,1],[40,0]] becomes 50x1 + 40x0 = 50
[[150,1],[4,1]] becomes 150x1 + 41x0 = 154 etc.

To summarize, a function like:

[[A,B],[C,D]] becomes AxB + CxD

answered Jul 27, 2021 at 15:26

Nicolas Martin

4,6091 gold badge6 silver badges15 bronze badges

$\begingroup$ This is a downsample task. I am trying to solve a bigger problem and for that we need a good prediction model so that it can predict sum of sequences based on flag value $\endgroup$
– Anup Patel
Jul 28, 2021 at 13:54
$\begingroup$ Understood. Have you tried a random forest? I could take different situations with flags, it could be a good solution. $\endgroup$
– Nicolas Martin
Jul 28, 2021 at 14:34

Add a comment |

1 Answer 1