No greed. No fear. Just maths.
Algorithmic Quant Trading (Machine Learning + Stat-Arb) 25 replies
Forex with Machine Learning Software Project 1 reply
Machine Learning + Retail Forex = Profitable? (Quant) 1 reply
Potential new machine learning style software. 79 replies
My most recent advancements into machine learning 16 replies
DislikedI like a lot the smoothing that the extreme learning nn gives maybe its usefull in noisy conditions just like financial times series did you tried to increase the number of hidden neurons ? to see how the approximation is finer grained there is a kind of GA build inside because of the random weights of the hidden layer i think the matrix is picking out the nodes that matter should see if some coefficients are zeroedIgnored
DislikedHow can I know if this forecast makes sense or if it is just plain random? If it is valid, how can I measure the predictive power of such an indicator? {image}Ignored
for(int i=barstoaction;i>=0;i--){ // Down Algo if( getRSI3(10,i,1.85,6)>44 ){ DrawArrow(string(objcounter)+"do",Time[i],High[i],242,3,clrRed,75*Point);objcounter++; if(i>=1){ bool tradecorrect=false; if(Open[i]>=Open[i-1]){ DrawLine(string(Time[i])+"dnco",Time[i],Close[i],Time[i],Close[i-1],2,clrLime);objcounter++; correct++; tradeswronginarow=0; balance+=0.8; }else{ DrawLine(string(Time[i])+"dnwr",Time[i],Close[i],Time[i],Close[i-1],2,clrRed);objcounter++; wrong++; tradeswronginarow++; if(tradeswronginarow>tradeswronginarowmax){tradeswronginarowmax=tradeswronginarow;} balance-=1; } } } } Comment(....balance...)
DislikedFrom what I found on the net Point gives the pip/point value of the current instrument. NormalizeDouble() is a rounding function and I guess int() returns the integral part of its operand. What does this normalization bring over simply using x[i]-y[i]?Ignored
Disliked- Why do you believe this is a good estimator for the worthfulness of this indicator?Ignored
Disliked- Do you consider a predictor which is right 75% of the time is good?Ignored
Disliked- x[i]-y[i] is the difference between the forecast and the actual value. Do you think 0 is a good value?Ignored
Disliked- Let's say I take a large number of samples and sum/average the values of x[i]-y[i]. Intuitively what do think the value will be?Ignored
Disliked- On the screenshot of my post the forecast is bearish. The target price is hit after only 2 days and again near the end of the forecasting period. At the forecast time the price is actually above the starting price. Do you consider the forecast is correct?Ignored
DislikedYou guys might want to check out my thread- http://www.forexfactory.com/showthread.php?t=528926 I have some really incredible ideas with a new ML algorithm, which isn't currently being used by any finance application forecasting-truly ground breaking stuff I am looking for a skilled coder to help me implement this idea, I will discuss more through PM for anyone who is serious. Thanks!Ignored
Disliked{quote} In your thread you say "some incredibly promising open-source software" and "using their proprietary algorithms". Either it's open or it's not. What is the license? If it is open source can you point us to their work to let us asses how incredibly promising it may be?Ignored
QuoteDisliked[...] leaves a lot to be answered for. For example the SL; originally it was 60% of ATR but later i remember you stating that an 80% SL worked better, but then you never really mentioned more about it??? did you ditch the 60% over the 80%? Other rules such as when to move SL are still fairly unclear. Do we move it on every new bar(and we're still in the trade!)? Move it once our same entry pattern re-appears? dont move it at all??? what about TP targets? is there a TP target? at one point i remember you discussing a 1.5 ATR as TP??? but again not...
QuoteDislikedWhat is our "pattern choosing" criteria? the pattern that shows up most in the last 200 bars?
QuoteDislikedWhat is our "pattern choosing" criteria?
1.13997, 1.13658, 1.13784, 1.13339, 1.13395, 1.13639, 1.12027, 1.11947, 1.11766, 1.11828, 1.10791, 1.10318, 1.08426, 1.08348, 1.07068, 1.05478, 1.06265, 1.04964
-33.9, 12.6, -44.5, 5.6, 24.4, -161.2, -8.0, -18.1, 6.2, -103.7, -47.3, -189.2, -7.8, -128.0, -159.0, 78.7, -130.1
-33.9 12.6 -44.5 12.6 -44.5 5.6 -44.5 5.6 24.4 5.6 24.4 -161.2 24.4 -161.2 -8 -161.2 -8 -18.1 -8 -18.1 6.2 -18.1 6.2 -103.7 6.2 -103.7 -47.3 -103.7 -47.3 -189.2 -47.3 -189.2 -7.8 -189.2 -7.8 -128 -7.8 -128 -159 -128 -159 78.7 -159 78.7 -130.1
QuoteDislikedI know you have touched on it before but can we have some more explanation as to why this does not work on other pairs? i mean the logic would dictate its a "learning" strategy, so it should "learn" the best "pattern" to trade regardless of pair selection? i understand that open and close times of the daily candles can effect this to some degree but would'nt you expect the "learning" to figure it out anyway? and find the best pattern. I ran a GU test and got horrible results...
The strategy has 4 variables belonging to the machine learning process. They are:
Using A = 0 (GMT +1/+2), B = 130, C = 9 and D = 12 we can obtain the result below for the EURJPY. Notice that:
DislikedGlad to see you back algoTraderJo! I'm following your thread with interest.Ignored