What’s the benefit to use Convolutional Neural Network?
There has been an optimization function to train parameters already.
Why do we need to use a similar function instead of a solution that we have been familiar with?
After all, a new function costs time to learn.
Good question.
Actually, several days ago, a user asked a similar question:
Please refer to that post.
Additionally, let me discuss more about the differences.
The optimizor is still good to train parameters for traditional algo trading.
But have to say that CNN is much better than it.
Start, End, Step are obsolete to train some parameters when the parameters are not based on linear distribution.
For example, if one parameter is a normal distribution with a mean of 0, then to test a value far from 0 is really not worth doing. Even if you can get the parameters fit the curve(equity), the generalization will be poor and there must be an over-fitting issue.
And, about what you said, the learning curve and learning cost, I think CNN is definitely worth learning. It’s not only for trading field, but you can apply it in any data science field. If you have experience to develop perceptron NN, CNN will not be that hard to learn. A week to a month is enough.
Fintechee has built-in plugins to support CNN, it’s just a tool out of box.
Yes. Thank you for your teaching!
I have learned how to build a new CNN according to the manual on your Github repo.
Now, I’m challenging how to train the model.