We are doing a project to detect ADHD from a video and we extract a feature vector from each frame. Now we want to extract a fixed length feature vector for the whole video and we think that we can use markov model in that purpose. Could you tell us if your code can be helpful in our case as we aren't sure how we could use it for training and testing.

Hi, I'm trying to use Auto_HMM to find upward and downward, and stable states of the stock's close prices time series. This model is supposed to learn to distinguish which period has an upward trend and which one has a downward trend and which hasn't a specific trend(rather monotonic). Below you can see the Adj Close Price data from S&P 500 index between '2010-07-01' and '2011-11-01'. Three periods are denoted by a line, each with a different color. I want to use this data to feed the HMM model and get the states according to periods. Here is what I have tried so far:

import pandas as pd
from pandas_datareader import data as web
import numpy as np

x = web.get_data_yahoo('^GSPC' , start = '2010-07-01' , end = '2011-11-01')['Adj Close'].rename('close')
x[250:].to_csv("the path")

# ------------------------------------------------------------------------------------------------
 # then, I used your code:
from Hidden_Markov_Model import *
from Hidden_Markov_Model.HMM import *
import time

Start=time.time()
Train_ratio=0.1
Cov_Type='diag'
Max_state=3
Max_mixture=2
Iter=1000
Feat=1

# I have 140 records, so I guess N should be equal to 140.
N=140
T=50
flag=1
Path="the path"
Data=pd.read_csv(Path)
Exam_HMM=Supervised_HMM(Train_ratio,Cov_Type,Max_state,Max_mixture,Iter,Feat,N,T,Data,flag)
Exam_HMM.Best_States()
END=time.time()
print('Total Time Takes in seconds',END-Start)

But I got this error:

One mixture component is over 1
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_7616/86438458.py in <module>
     17 Data=pd.read_csv(Path)
     18 Exam_HMM=Supervised_HMM(Train_ratio,Cov_Type,Max_state,Max_mixture,Iter,Feat,N,T,Data,flag)
---> 19 Exam_HMM.Best_States()
     20 END=time.time()
     21 print('Total Time Takes in seconds',END-Start)

c:\Users\Shayan\Desktop\Hidden_Markov_Model\HMM.py in Best_States(self)
    146         self.Len=[self.T for ii in range(0,self.Data_train.shape[0])]   # Lengths must be list
    147         self.Train_Data = np.array(self.Data_train).reshape((-1,1)) # Convert to numpy array with one column
--> 148         self.AIC_BIC()
    149         self.Best_BIC()
    150 

c:\Users\Shayan\Desktop\Hidden_Markov_Model\HMM.py in AIC_BIC(self)
    176             # self.num_params=self.Max_state*(self.Max_state-1)+ self.Max_state*(ii-1)+(ii*self.Max_state)*self.Feat+((self.Feat**2+self.Feat)/2)*ii*self.Max_state  # Full Covariance
    177             self.num_params = self.Max_state*(self.Max_state-1)+ self.Max_state*(ii-1)+(ii*self.Max_state)*self.Feat+(self.Max_state*ii*self.Feat)  # Diagonal
--> 178             Model=GMMHMM(n_components=self.Max_state,n_mix=ii,covariance_type=self.Cov_Type,params='stmcw', init_params='stmcw',tol=pow(10,-5),n_iter=self.Iter).fit(self.Train_Data,self.Len)
    179             AIC.append(-2 * Model.score(self.Train_Data) + 2 * self.num_params)
    180             BIC.append(-2 * Model.score(self.Train_Data) +  self.num_params * np.log(self.Train_Data.shape[0]))

~\Anaconda3\envs\Python3.10\lib\site-packages\hmmlearn\base.py in fit(self, X, lengths)
    504             stats = self._initialize_sufficient_statistics()
    505             curr_log_prob = 0
--> 506             for sub_X in _utils.split_X_lengths(X, lengths):
    507                 lattice, log_prob, posteriors, fwdlattice, bwdlattice = \
    508                         impl(sub_X)

~\Anaconda3\envs\Python3.10\lib\site-packages\hmmlearn\_utils.py in split_X_lengths(X, lengths)
     14         n_samples = len(X)
     15         if cs[-1] > n_samples:
---> 16             raise ValueError(
     17                 f"more than {n_samples} samples in lengths array {lengths}")
     18         elif cs[-1] != n_samples:

ValueError: more than 14 samples in lengths array [50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50, 50]

Can you please help me? How can I achieve my goal?

The expected output in my case:

[0,0,0,0,0,0,0...,0,1,1,1,1,1,1,1,1,...,1,2,2,2,2,2,2,2...,2]

I expect HMM to find those three states(downward trend - monotonic period - upward trend).

Hi Manitadayon, I am glad to find your code and thanks from sharig it. I am new in machine learning. Now I want to code for this paper and want to use this dataset. But i don't know how can i do this. Can you help me? I don't know exactly what is states in this paper and how give dataset to your code, and how wite code for this paper.