Machine learning in trading: theory, models, practice and algo-trading - page 928

[Mihail Marchukajtes]

Maxim Dmitrievsky:

The main thing is that Akello won't miss again next week.

Thought my basic strategies sucked. Can someone give me some basic strategies, which I will try to improve with my agents?

Take a counter-trend. I'm telling you as a surgeon. It's the most appropriate approach. There's no better time for analysis than the beginning of a pullback....

[Maxim Dmitrievsky]

Mihail Marchukajtes:

Take a countertrend. I'm telling you as a surgeon. It's the most appropriate approach. There's no better time for analysis than the beginning of a pullback....

https://www.mql5.com/en/code/19598

I'll take this one for now, it's an ancient strategy, but it's good for optics

[Mihail Marchukajtes]

The basic strategy is needed ONLY to select the moment (time) for the analysis. It can be static and have no optimization parameters. If we optimize the basic strategy, we get a shitload of models. There is no point in optimizing the basic strategy. The loading is taken by the NS. It is enough to set a convenient set of parameters in the basic strategy, in terms of number of deals per day, and already use it to train NS....

[Dr. Trader]

mnogovhodov_02 2016 arr_Buy turned out like this:

	y_pred
y_true	0	1
0	101797	52445
1	24310	24208

[Mihail Marchukajtes]

Maxim Dmitrievsky:

The girl says you're right, but we need more champagne.

She's cool, tell her I said hi!!!!

[Maxim Dmitrievsky]

Mihail Marchukajtes:

Cool, tell her I said hi!!!!

OK, you too from her :)

[Dr. Trader]

Alternative. The result immediately in classes, without probabilities. That seems worse to me.

	y_pred
y_true	0	1
0	81744	72498
1	18618	29900

[Mihail Marchukajtes]

Maxim Dmitrievsky:
OK, you too from her :)

Maybe it will teach you some sober thoughts about TC. How you can and can't do.....

[Maxim Dmitrievsky]

Mihail Marchukajtes:

Maybe it will teach you some sober thoughts about TC. How you can and can't do.....

She says I'm smart and just have to take her to Australia, she has a friend there

We need a fake marriage for that

[Dr. Trader]

SanSanych Fomenko:

I stopped counting errors by these tables as standard.

My reasoning is, I take class 1 right away, it is more obvious: Initial class "0" gave class "1" prediction = 86118, and class "1" gave class "1" prediction = 12256. This means that when trading, we will get false class predictions = 86118, while correct predictions = 12256, i.e. error = 86116/(86116+12256) = 87.5%9(!!), if class "1" = entry/position, then it is a disaster. But the position of class "0" is very decent - the erroneous zeros in decision making will be only 5.3%.

I don't use such tables when comparing models, either. I added it here just for clarity. You can see right away that there are a lot of zeros predicted where there should be "1", and you can see that with one tree the forecast is bad.