What kind of neural network i should use?

this question is in relation to my previous question regarding control of laser pointer using two actuator movement.I am finally not getting consistent results in every training data.Is there something i should do with my training data.I move my actuators one by one some sample training data is given here.For that actuator movement i know laser pointers motion.Now i feed network with laser pointers movement as input and actuator motion as output.(The reverse of what actually happens).Because i want to know actuator position for laser pointer at origin. The input data is given below:

%% input is

• laser point_x laser point_y
• -3.9976E-06 -508.2157873
• -168.2815517 -377.9060761
• -125.9567908 -246.798648
• -126.9634896 -285.1190368
• -144.0240458 -323.4315425
• -161.0842613 -361.7432835
• -178.1441362 -400.0542598
• -195.2036706 -438.3644714
• -212.2628643 -476.6739184
• -229.3217175 -514.9826007
• -246.38023 -553.2905183
• -263.4384019 -591.5976713
• -280.4962333 -629.9040595
• -297.553724 -668.2096831
• -314.6108742 -706.514542
• -331.6676837 -744.8186362
• -348.7241526 -783.1219658
• -365.780281 -821.4245307
• -382.8360687 -859.7263309
• -399.8915159 -898.0273664

Outputs are z1:

• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859
• -25048.71859and

z2

• -25042.23525
• -25041.23525
• -25040.23525
• -25039.23525
• -25038.23525
• -25037.23525
• -25036.23525
• -25035.23525
• -25034.23525
• -25033.23525
• -25032.23525
• -25031.23525
• -25030.23525
• -25029.23525
• -25028.23525
• -25027.23525
• -25026.23525
• -25025.23525
• -25024.23525
• -25023.23525

using this data if any one could help me frame a network that gives output z1=-25028.17074 and z2=-25022.99735 these values are output for input laser position (-6.20458E-12,-6.49523E-11)=(0,0) Thanks in advance.please please any experienced professor guide me through this big hurdel.please!!!!

[Greg Heath]

If you are going to post data without an attachment, it would be appreciated if the data can just be copied directly from the post and pasted into a text file.

[Greg Heath]

z1 has only 19 points and they are all equal

Did you really want 20 equal z1 points? i.e. a vertical line in the z plane?

[sameer d]

To make it easy for the network,i tried keeping z1 constant and make z2 only to vary.i tested if the network could (for gods sake) predict z2 correctly!but that also didnt happen.plz guide me.as of actual case z1 and z2 both would vary.thanks

Answer 1

[Greg Heath]

I have already answered this question recently. Please reread it.

 1. You need 2 nets trained on the same data.
 2. A = net1(L) and the inverse L = net2(A).

Thank you for formally accepting my answer

Greg