| Title: | Hidden Markov Model for Predicting Time Sequences with Mixture Sampling |
|---|---|
| Description: | An algorithm for time series analysis that leverages hidden Markov models, cluster analysis, and mixture distributions to segment data, detect patterns and predict future sequences. |
| Authors: | Giancarlo Vercellino [aut, cre, cph] |
| Maintainer: | Giancarlo Vercellino <[email protected]> |
| License: | GPL-3 |
| Version: | 1.0.2 |
| Built: | 2025-01-09 07:01:34 UTC |
| Source: | CRAN |
A data frame with the close prices for AMZN, NVDA and IBM
dummy_setdummy_set
An object of class data.frame with 1925 rows and 4 columns.
Yahoo Finance
An algorithm for time series analysis that leverages hidden Markov models, cluster analysis, and mixture distributions to segment data, detect patterns and predict future sequences.
hmix function segments the time series with k-means clustering, fits an HMM to model state transitions, and generates future predictions over a specified horizon. It evaluates model accuracy by calculating the Continuous Ranked Probability Score (CRPS) across multiple test points, producing error metrics that assess the model's predictive performance and robustness.
hmix( ts, horizon, centers = 10, n_hidden = 4, seed = 42, n_tests = 20, warmup = 0.5 )hmix( ts, horizon, centers = 10, n_hidden = 4, seed = 42, n_tests = 20, warmup = 0.5 )
ts |
A numeric vector representing the time series data. |
horizon |
Integer. The prediction horizon, specifying how many future points to forecast. |
centers |
Integer. Number of clusters for k-means clustering. Default: 10. |
|
Integer. Number of hidden states in the Hidden Markov Model. Default: 4. |
|
seed |
Integer. Random seed for reproducibility. Default: 42. |
n_tests |
Integer. Number of testing points for back-testing. Default: 20. |
warmup |
Numeric. Proportion of the time series used as the warm-up period before testing. Default: 0.5. |
This function returns a list containing:
model: The HMM model along with its estimated parameters.
hmm_model: The object includes classified observations, initial HMM and trained HMM.
pred_funs: Prediction functions for each point in horizon (rfun, dfun, pfun, qfun)
error_sets: A list of error metrics calculated for each testing point (at this time, CRPS).
Maintainer: Giancarlo Vercellino [email protected] [copyright holder]
Useful links:
# Example usage of hmix function: result <- hmix(dummy_set$AMZN, horizon = 10, centers = 5, n_hidden = 3, n_tests = 2) print(result$model) print(result$error_sets) # Random sampling for each point in predicted horizon result$model$pred_funs$t1$rfun(10) # ICDF for each point in horizon result$model$pred_funs$t5$qfun(c(0, 1)) # PDF for each point in horizon result$model$pred_funs$t8$dfun(tail(ts)) # CDF for each point in horizon result$model$pred_funs$t10$pfun(tail(ts))# Example usage of hmix function: result <- hmix(dummy_set$AMZN, horizon = 10, centers = 5, n_hidden = 3, n_tests = 2) print(result$model) print(result$error_sets) # Random sampling for each point in predicted horizon result$model$pred_funs$t1$rfun(10) # ICDF for each point in horizon result$model$pred_funs$t5$qfun(c(0, 1)) # PDF for each point in horizon result$model$pred_funs$t8$dfun(tail(ts)) # CDF for each point in horizon result$model$pred_funs$t10$pfun(tail(ts))