Machine learning in trading: theory, models, practice and algo-trading - page 3742

New comment
 
Two identical locally deployed LLMs (run on their own hardware) will give matching answers to the same chain of first questions?
 
fxsaber #:

Am I right that if I lost an ONNX file with a model, there is no way to recover it? Even if the data that allowed to create this ONNX is completely saved?

Roughly speaking, you can't get two identical models.

Tree models of MO use pseudo-random or really random mixing of rows, columns.
Usually there is a parameter Seed which sets the initial value of pseudo-random HGF, if Seed was not changed at new training, there can be repetitions (I can't say about all packages/software, but forest on the basis of alglib gave identical models. But it is recommended to set Seed for example from current time. Other MO programmes can, for example, by omission use time or use really random HSCs. There are already thousands of them and who coded what in his programme - you need to read the manual.
I have forgotten about neuroncs, but it seems that somewhere random is used there too. Unchanged Seed will probably create a copy of the model.
fxsaber #:
Two identical locally deployed LLMs (run on their own hardware) will give the same answers to the same chain of first questions?
I don't use LLMs, so I don't know. Perhaps someone else knows.
 
Aleksei Kuznetsov #:
Tree models of MO use pseudo-random or really random mixing of rows, columns.
Usually there is a parameter Seed that sets the initial value of pseudo-random HGF, if Seed has not changed during new training, there can be repetitions (I can't say about all packages/software, but the forest based on alglib gave the same models. But it is recommended to set Seed for example from current time. Other MO programmes can, for example, by omission use time or use really random HSCs. There are already thousands of them and who coded what in his programme - you need to read the manual.
I have forgotten about neuroncs, but it seems that somewhere random is used there too. Unchanged Seed will probably create a copy of the model. I don't use LLM, so I don't know. Maybe someone else knows.
If even GA doesn't allow to set Seed (values change from run to run), what to speak about learning algorithms....
 
fxsaber #:
If even GA does not allow to set Seed (values change from run to run), what to say about learning algorithms....
Some of them do - you should check the documentation of a particular programme.
1...37353736373737383739374037413742
New comment