train.py

from data import BVDataset
from LDSModel import LDS
import pickle
import numpy as np
import random
np.random.seed(420)
random.seed(420) 




if(0):
    dataSet = BVDataset()
    ch=range(0,100,10)
    all_train_data = dataSet.getData(ch=ch)
    dataSet.plot(all_train_data,title='train-good-data')

    # Data to be saved
    data = {
            'all_train_data': all_train_data       
        }

        # Save data to a file
    with open('all_data.pkl', 'wb') as f:
        pickle.dump(data, f)

# Load data from the file
with open('all_data.pkl', 'rb') as f:
    loaded_data = pickle.load(f)

all_train_data = loaded_data['all_train_data']





train_data_mean=np.mean(all_train_data)
train_data_var=np.var(all_train_data)

# #model = LDS(mu=np.mean(all_train_data),sigma=np.var(all_train_data))
# # Initialize the Kalman filter Model
# n_states=1
# transition_matrix_guess = np.random.rand(n_states, n_states)
# observation_matrix_guess = np.random.rand(1, n_states)
# initial_state_mean_guess = train_data_mean#np.random.rand(n_states)
# initial_state_covariance_guess = train_data_var#np.eye(n_states) * 0.1
# observation_covariance_guess = np.eye(1) * 0.1
# transition_covariance_guess = np.eye(n_states) * 0.1
# Create Kalman Filter

model = LDS(n_states=1)
model.make_initial_guess(all_train_data)
model.EM_init(all_train_data)




# Fit the model using EM algorithm on training data
loglikelihood_old = float('-inf')
loglikelihood_new,_,_ = model.loglikelihood(all_train_data)

stop_diff=0.001
max_iter=100
num_iter=0
while abs(loglikelihood_new - loglikelihood_old) > stop_diff:  # Convergence criterion
    loglikelihood_old = loglikelihood_new
    
    
    #model = model.em(all_train_data, n_iter=1,em_vars='all')

    model.EM(all_train_data,total_iterations=1)

    loglikelihood_new,_,_ = model.loglikelihood(all_train_data)
        
    num_iter=num_iter+1
    print(f"iter_{num_iter}, Log-likelihood: {loglikelihood_new}")
    if(num_iter>max_iter):
        break


# Save the Kalman filter object to a file
with open('LDS_Model.pkl', 'wb') as f:
    pickle.dump(model, f)