"Neural Market Trends Tutorial 2 problem"

sunnyfunghy

Hi everyone,

I would like to ask Neural Market Trends Tutorial 2 Part 1 and 2, using default mode for prediction. When I change with copying some training data, those GC trends (label) are all ("UP") into testing data, the result still all "Down" from the computer prediction. Moreover, when copying all the training data into testing data, the result are still predict to "Down". Actually, I follow the steps of tutorials. But the result seems that unreasonable. Can anyone explain why? Does anyone notice? Thank you very much


Cheers,
Sunny

fritmore

sunnyfunghy wrote:

Hi everyone, I would like to ask Neural Market Trends Tutorial 2 Part 1 and 2, using default mode for prediction. When I change with copying some training data, those GC trends (label) are all ("UP") into testing data, the result still all "Down" from the computer prediction. Moreover, when copying all the training data into testing data, the result are still predict to "Down". Actually, I follow the steps of tutorials. But the result seems that unreasonable. Can anyone explain why? Does anyone notice? Thank you very much Cheers, Sunny

hi Sunny
Well i havent played with that tut.
Maybe the model simply predicts wrongly?

if you used the same training as the testing data you should get the same result/predictions for both sets (at least for most of the model paradigms I can think of).

sunnyfunghy

I connect training data into default model and use apply model to test the testing data to test the result. If wrong, how do I fix it?

IngoRM

Hi,

and there you have the reason: The default model always predicts the major class. It can be used as baseline but is not recommended as a real learning scheme. Try and replace the default learner by a different learning scheme which actually learns something.

Cheers,
Ingo