# Copyright 2024, MetaQuotes Ltd.
# https://www.mql5.com
from hmmlearn import hmm
import numpy as np
import pandas as pd
import json 


assumed_states = 2 #number of states of process
maxhmm_iters = 10000 #maximum number of iterations for optimization procedure

def hmm2json(hmm_model, filename):
    """
    function save a GaussianHMM model to json format 
    readable from MQL5 code.
    param: hmm_model should an instance of GaussianHMM
    param: string. filename or path to file where HMM 
    parameters will be written to
    """
    if hmm_model.__class__.__name__ != 'GaussianHMM':
        raise TypeError(f'invalid type supplied')
    if len(filename) < 1 or not isinstance(filename,str):
        raise TypeError(f'invalid filename supplied')
    jm  = {
            "numstates":hmm_model.n_components,
            "numvars":hmm_model.n_features,
            "algorithm":str(hmm_model.algorithm),
            "implementation":str(hmm_model.implementation), 
            "initprobs":hmm_model.startprob_.tolist(),
            "means":hmm_model.means_.tolist(),
            "transitions":hmm_model.transmat_.tolist(),
            "covars":hmm_model.covars_.tolist()
          }
    with open(filename,'w') as file:
        json.dump(jm,file,indent=None,separators=(',', ':')) 
    return 
#dataset to train model on    
dataset = np.array([[0.56807844,0.67179966,0.13639585,0.15092627,0.17708295],
                   [0.62290044,0.15188847,0.91947761,0.29483647,0.34073613],
                   [0.47687505,0.06388765,0.20589139,0.16474974,0.64383775],
                   [0.25606858,0.50927144,0.49009671,0.0284832,0.37357852],
                   [0.95855305,0.93687549,0.88496015,0.48772751,0.10256193],
                   [0.36752403,0.5283874 ,0.52245909,0.77968798,0.88154157],
                   [0.35161822,0.50672902,0.7722671,0.56911901,0.98874104],
                   [0.20354888,0.82106204,0.60828044,0.13380222,0.4181293,],
                   [0.43461371,0.60170739,0.56270993,0.46426138,0.53733481],
                   [0.51646574,0.54536398,0.03818231,0.32574409,0.95260478]])  
#instantiate an HMM and train on dataset
model = hmm.GaussianHMM(assumed_states,n_iter=maxhmm_iters,covariance_type='full',random_state=125, verbose=True).fit(dataset)  
#save the model to the common folder of Metatrader 5 install
hmm2json(model,r'C:\Users\Zwelithini\AppData\Roaming\MetaQuotes\Terminal\Common\Files\exampleHMM.json')
#get the state probabilities and log likelihood
result = model.score_samples(dataset)
print("log_likelihood " ,result[0]) #print the loglikelihood
print("state sequence ", model.decode(dataset)[1]) #print the state sequence of dataset
print("state probs ", result[1]) #print the state probabilities