SOM: cooking methods - page 7
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Made his version, while preliminarily on the OOS plum, though on 100% will not claim - the first test yet.
In this context, there are a few questions:
1. I'm getting more than an order of magnitude discrepancy in the number of dropouts, I don't think this is good. I have some thoughts on how to fix it, but not 100%. Can anyone suggest something.
2. to the topic-starter:
On what instruments is there a positive result on the OOS? The TF D1 is a lot, we need at most H4 to get more or less quality statistics. If you have a desire to feel, drop your soapbox to me in my personal message. I'm not ready to put it on the market yet, so only for your personal use.
The Expert Advisor uses a dll -- there are two ways -- check the online dll for viruses or compile it, you can choose
--- 1. I'm getting more than an order of magnitude discrepancy in the number of dropouts, I don't think that's good. I have some thoughts on how to fix it, but not 100%. Can someone suggest something.
I don't understand what the dropouts mean and what the discrepancies are with. With excel?
--- 2. Topikstarter:
On what instruments is there a positive result on OOS? The TF D1 is a lot, we need at most H4 to get more or less quality statistics. If you have a desire to feel, drop your soapbox to me in a personal message. For now I am not ready to put it on the market, so only for your personal use.
What strategy did you use to implement the Expert Advisor? Bought and kept the specified number of bars?
I got a good result on D1. I have not tested it on H4, on H1 it needs more time to check, but the fact is that the probability of price movement is not very different from 50%. Most likely my original strategy is best suited for large TF.
I will throw the soap, look, compare with my Expert Advisor (my Expert Advisor template with binding to dll has been written by programmer for a fee).
I don't understand what the drop-downs mean and what the discrepancies are with. With Excel?
My bad.
Fallouts are the number of input patterns that hit the cluster of a particular neuron. The terminology is mine :) I made it up on the fly.
The discrepancies between the number of falls -- on different neurons -- i.e. one has, say, 15 falls, the other has 400 -- are not right, imho. That's what I meant.
Now I want to try to take it into account in training.
What strategy did you use to implement your Expert Advisor? Bought and holds a given number of bars?
No, all the time in the market. Actually yes, need to tweak the strategy.
With this strategy I've got a good result on D1. I haven't tested it on H4, on H1 it needs more time, but the fact is the probability of price movement is not very different from 50%. Probably my original strategy is best suited for large TF.
Hm, did the math. The minimum requirement for the grid is about 1000 inputs, i.e. about 4 years are thrown away, leaving 7 years of OOS for statistics, I will have to check.
Well, then I'll upload it tomorrow at the earliest. I also need to get the training setting in the Expert Advisor out of the dll.I see now.
I trained the network on clock bars for 10 years, i.e. a little more than 60 thousand examples.
Here is the distribution of examples per neuron (100 neurons), X-axis is neuron numbers, Y-axis is number of examples.
Once I got very uneven distribution, when I tried to train SOM on candlesticks information (all 4 prices of candlesticks were taken into account). Apparently some candlestick patterns are much more common than others. The map showed an obvious cluster with a big vertex and all the other neurons with few cases.
--- Hmmm, did the math. The minimum requirement for the grid is about 1000 inputs, so that's about 4 years thrown away, leaving 7 years of OOS for statistics, I'll have to check.
How did you calculate that? IMHO there is overkill from the size of the input vector.
How did you calculate this? IMHO there is an overshoot from the size of the input vector.
Eurobucks diaries from 2004 to now:
The Eurobucks diary report from 2004 to now:
Can I see the report (header) in full? The bottom part is cut down.