Mt4 Probability Script - python-2.7

Question

I'm fairly new to python
I have made a simple script that imports price feeds from mt4
My idea / Project is to turn this into some sort of a probability indicator, that is giving the probability, besides the bid and ask,
for example:
TIME/ BID ASK
USD/CADD 22:19 1.30451 60%^ 1.30D39 40%v
and the probability is changing within specific period, i.e for example 1hr period, so every hour it will give a new probability of the direction
It is looking for two patterns: A, B,
Pattern A represents a bullish pattern
Pattern B represents a bearish pattern
basically looking for how strong is the probability A or B reoccurring
out of the two which has a higher chance of reoccurring,
Here is where I am stuck
I have no idea how to put that together...
Here is what I have so far:
import datetime
import numpy as np
import pandas as pd
import sklearn
from pandas.io.data import DataReader
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.lda import LDA
from sklearn.metrics import confusion_matrix
from sklearn.qda import QDA
from sklearn.svm import LinearSVC, SVC
import dde_client as ddec
import time
QUOTE_client = ddec.DDEClient('MT4', 'QUOTE')
symbols = ['USDCAD', 'GOLD','EURUSD', 'SILVER', 'US30Cash', ]
for i in symbols:
QUOTE_client.advise(i)
def Get_quote():
while 1:
time.sleep(1)
print "Symbol\tDATE\t\tTIME\tBID\tASK"
for i in symbols:
current_quote = QUOTE_client.request(i).split(" ")
day, _time, bid, ask = (current_quote[0], current_quote[1],
current_quote[2], current_quote[3])
print i + ":\t" + day + "\t" + _time +"\t" +bid+ "\t" + ask
break
time.sleep(1)
return Get_quote()
continue
def create_lagged_series(cuurent_quote,start_time, end_time, lags=1):
ts = DataReader(cuurent_quote,symbols,
start_time-datetime.timedelta(hours=1),
end_time
)
tslag = pd.DataFrame(index=ts.index)
ts['Today'] = ts['Adj Close']
tslag["Volume"] = ts["Volume"]
for i in xrange(0, lags):
tslag["Lag%s" % str(i+1)] = ts["Adj Close"].shift(i+1)
tsret = pd.DataFrame(index=tslag.index)
tsret["Volume"] = tslag["Volume"]
tsret["Today"] = tslag["Today"].pct_change()*100.0
for i in xrange(0, lags):
if (abs(x) < 0.0001):
tsret["Today"][i] = 0.0001
for i in xrange(0,lags):
tsret["Lag%s" % str(i+1)] = \
tslag["Lag%s" % str(i+1)].pct_change()*100.0
tsret["Direction"] = np.sign(tsret["Today"])
tsret = tsret[tsret.index >= start_time]
return tsret
if __name__ == "__main__":
snpert = create_lagged_series(len('GOLD', Get_quote(), 1))
X = snpret[["Lag1","Lag2"]]
y = snpret["Direction"]
start_test = cuurernt_quote
X_train = X[X.index < start_test]
X_test = X[X.index >= start_test]
y_train = y[y.index < start_test]
y_test = y[y.index >= start_test]
print "Hit Rates/Confusion Matrices:\n"
models = [ ( "LR", LogisticRegression() ),
( "LDA", LDA() ),
( "QDA", QDA() ),
( "LSVC", LinearSVC() ),
( "RSVM", SVC( C = 1000000.0,
cache_size = 200,
class_weight = None,
coef0 = 0.0,
degree = 3,
gamma = 0.0001,
kernel = 'rbf',
max_iter = -1,
probability = False,
random_state = None,
shrinking = True,
tol = 0.001,
verbose = False
)
),
( "RF", RandomForestClassifier( n_estimators = 1000,
criterion = 'gini',
max_depth = None,
min_samples_split = 2,
min_samples_leaf = 1,
max_features = 'auto',
bootstrap = True,
oob_score = False,
n_jobs = 1,
random_state = None,
verbose = 0
)
)
]
# Iterate through the models
for m in models:
# Train each of the models on the training set
m[1].fit(X_train, y_train)
pred = m[1].predict(X_test)
print "%s:\n%0.3f" % (m[0], m[1].score(X_test, y_test))
print "%s\n" % confusion_matrix(pred, y_test)
Here is just my MT4 price feed script on its own:
import dde_client as ddec
import time
__author__ = 'forex Ticker'
print __author__
QUOTE_client = ddec.DDEClient('MT4', 'QUOTE')
symbols = ['USDCAD', 'GOLD','EURUSD', 'SILVER', 'US30Cash', ]
for i in symbols:
QUOTE_client.advise(i)
while 1:
time.sleep(1)
print "Symbol\tDATE\t\tTIME\tBID\tASK"
for i in symbols:
current_quote = QUOTE_client.request(i).split(" ")
day, _time, bid, ask = (current_quote[0], current_quote[1],
current_quote[2], current_quote[3])
print i + ":\t" + day + "\t" + _time +"\t" +bid+ "\t" + ask
break
time.sleep(1)
continue

Answer 1

As to fixing your algorithm I suggest that rather than doing it from scratch and hacking various libraries together, start from a working example and modify it to your liking. You don't have to understand it completely, but you need something to start with.
I would even abstract away the MT4 and quotes reading logic, and just have some test numbers (fake or sampled) in a CSV or TXT file. Start form a working example that can recognize A and B patterns in this file. Having that define how your own algorithm is different and try to adjust it.
When it's working the final step is the MT4 integration. It looks like the DDE server is only meant for data export, but not for building an indicator. Consider this alternative framework: linking MT4 to Python. Not only you can build an on-chart indicator with it, but also perform automatic trading using your algorithm.

Answer 2

Q: How to put that together...? A: Have a Realistic Plan - best before one puts the money on table. That can save youfrom even starting to do a nonsense orfrom aiming at non-realistic targets.
No one would be harmed, if plan is the first working document elaborated and agreed by all parties involved on HOW the "great and cool new disruptive vision" WILL BE CREATED.
Organise your further work in steps + always add budget controls, be it in [man*weeks] or [k$], one is willing to spend on items.
One ought be able to decide on the feasibility and surviveability of the initially great & cool idea.
Plan carefully inside the principal phases, both on the MQL4/5 side, python and other components:
X [man*weeks] on [System Integration Architecture],
Y [man*weeks] on [Integration Model Design],
Z [man*weeks] on [Integration Model Prototype],
U [man*weeks] on [Integration Model Testing],
V [man*weeks] on [Integration Model Release],
W [man*weeks] on [Integration Model Production Ecosystem]
S [man*weeks] on [Design Cycles on finding best Predictions Model]
T [man*weeks] on [Design Cycles on finding good Trading Strategies for Predictions]
Items not to be forgotten to overcome in the early architecture decisions:
0) Forget to use MQL4/5 examples, you put yourself at risk into sub-millisecond domain battle with hundreds of millions USD in fight and motion
1) Forget to use Custom Indicator in MQL4/5 MetaTrader Terminal ( blocking )
2) Forget to use DDE integration, some O/S do not support it at all
3) Forget to use pandas ( even for any AI/ML-model prototyping ) as nanoseconds matter a lot in ML-process, pandas is a great toy, but not for the performance a real-world trading needs for ML-model tuning.
4) Forget to use start-end logic, the AI/ML-engines have to be separate, in order to efficiently train/validate/test for their best generalisation abilities in vast HyperPARAM state-spaces.for m in models: can be in the source-code, but not in the reality. One instrument may take ( and does take ) about a few tens of [CPU-core*days] runtimes in parameter optimisation on COTS-hardware, so count with realistic numbers here, for proper budgeting of each of the [S]+[T] cycles.
Epilogue:
Anyway a smart Programme, if approved as financially feasible. May like other posts on Low-latency MT4-AI/ML-integration for algorithmic-trading.