HMM-XGB-Strategy

This is the HMM-XGB strategy from my internship in Taoli. It contains the whole Gaussian-HMM with XGBoost training and testing code.

Obtaining the data (using IH for example)
Using HMM to predict the hidden states of each data (remember change the number of hidden states)
Using XBGoost to retraining and repredicting, updating the hidden states if the log-likelihood improve.
Select time area to slide update model.

Using train_HMM_XGB_model.py to run the entire project.

Pay attention to project location, data location, running time, training or testing.

(Some parameters and hyper-parameters must be changed in own situation.)

For example, the hidden states, the hyper-parameters in XGBoost. And the raw_da.csv is an outdated example.

Actually, I also tried LSTM to exchange the XGBoost, but it needs more training time and data form changes(sequence form).

Name		Name	Last commit message	Last commit date
Latest commit History 4 Commits
GMM_HMM.py		GMM_HMM.py
HMM_XGB.py		HMM_XGB.py
LICENSE		LICENSE
README.md		README.md
SQLData.py		SQLData.py
data_code.py		data_code.py
form_B_matrix_by_XGB.py		form_B_matrix_by_XGB.py
pred_prob_HMM_XGB.py		pred_prob_HMM_XGB.py
raw_da.csv		raw_da.csv
re_estimate.py		re_estimate.py
return_back_test.py		return_back_test.py
self_pred.py		self_pred.py
self_xgb.py		self_xgb.py
train_HMM_XGB_model.py		train_HMM_XGB_model.py