# Copyright 2024, MetaQuotes Ltd.
# https://www.mql5.com
# imports
import MetaTrader5 as mt5
import numpy as np
import pandas as pd
from JsonModel import lfspy2json, LocalFeatureSelection, MQL5_COMMON_FOLDER, MQL5_FILES_FOLDER
from os import path
from sklearn.metrics import accuracy_score, classification_report

#initialize MT5 terminal
if not mt5.initialize():
    print("MT5 initialization failed ")
    mt5.shutdown()
    exit() # stop the script if mt5 not initialized
  
#we want to get the path to the MT5 file sandbox
#initialize TerminalInfo instance    
terminal_info = mt5.terminal_info()

#model file name
filename = "lfsmodel.json"

#build the full path
modelfilepath = path.join(terminal_info.data_path,MQL5_FILES_FOLDER,filename)

#number of random numbers to generate
datalen = 1000

#number of features the dataset will have
datavars = 5

#set random number seed
rng_seed = 125
rng = np.random.default_rng(rng_seed)

#generate the numbers
data = rng.uniform(-1.0,1.0,size=datalen)

#shape our dataset
data = data.reshape([datalen//datavars,datavars])

#set up container for class labels
class_labels = np.zeros(shape=data.shape[0],dtype=np.uint8)

#set the class labels
for i in range(data.shape[0]):
    class_labels[i] = 1 if (data[i,1] > 0.0 and data[i,2] > 0.0) or (data[i,1] < 0.0 and data[i,2] < 0.0) else 0

#partition our data
train_size = 100
xtrain = data[:train_size,:]
ytrain = class_labels[:train_size]

#load testing data (out of sample)
test_data = data[train_size:,:]
test_labels = class_labels[train_size:]

#here we prepare the out of sample data for export using pandas
#the data will be exported in a single csv file
colnames = [ f"var_{str(col+1)}" for col in range(test_data.shape[1])]
testdata = pd.DataFrame(test_data,columns=colnames)

#the last column will be the target labels
testdata["c_labels"]=test_labels

#display first 5 samples
print("Out of sample dataframe head \n", testdata.head())
#display last 5 samples
print("Out of sample dataframe tail \n", testdata.tail())

#build the full path of the csv file
testdatafilepath=path.join(terminal_info.data_path,MQL5_FILES_FOLDER,"testdata.csv")

#try save the file
try:
    testdata.to_csv(testdatafilepath)
except Exception as e:
    print(" Error saving iris test data ")
    print(e) 
else:
    print(" test data successfully saved to csv file ")  

#initialize the LFS object
lfs = LocalFeatureSelection(rr_seed=rng_seed,alpha=8,tau=2,n_beta=20,nrrp=2000)

#train the model
lfs.fit(xtrain,ytrain)

#get the inclusion matrix
fstar = lfs.fstar

#add up all ones for each row of the inclusion matrix
bins = fstar.sum(axis=1)

#calculate the percent of times a candidate was selected
percents = 100.0 * bins.astype(np.float64)/np.float64(ytrain.shape[0])
index = np.argsort(percents)[::-1]

#output the results
print("------------------------------> Percent of times selected <------------------------------" )
for i in range(percents.shape[0]):
   print(f" Variable  {colnames[index[i]]}, selected {percents[index[i]]} %")


#conduct out of sample test of trained model
accuracy = lfs.score(test_data,test_labels)
print(f" Out of sample accuracy is {accuracy*100.0} %")


#export the model
try:
    lfspy2json(lfs,modelfilepath)
except Exception as e:
    print(" Error saving lfs model ")
    print(e)
else:
   print("lfs model saved to \n ", modelfilepath)