-
Notifications
You must be signed in to change notification settings - Fork 14
/
README.Rmd
223 lines (152 loc) · 9.34 KB
/
README.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
---
output: github_document
bibliography: references.bib
---
<!-- README.md is generated from README.Rmd. Please edit that file -->
```{r, echo=FALSE, message=FALSE, warning=FALSE}
knitr::opts_chunk$set(
collapse = TRUE,
dpi = 300,
out.width = "100%",
comment = "#>",
fig.path = "man/figures/README-"
)
library(gmwm)
```
[](https://travis-ci.org/SMAC-Group/gmwm)
[](http://www.repostatus.org/#active)
[](https://opensource.org/licenses/AGPL-3.0)
[](https://cran.r-project.org/)
[](http://www.r-pkg.org/pkg/gmwm)
[](http://www.r-pkg.org/pkg/gmwm)
[)`-yellowgreen.svg)](https://github.com/SMAC-Group/gmwm)
# `gmwm` R Package <a href="https://data-analytics-lab.net/"><img src="man/figures/logo.png" align="right" alt=" " width="230"></a>
This repository holds the Generalized Method of Wavelet Moments (GMWM) R package. This estimation technique was introduces in @guerrier2013wavelet and uses the wavelet variance in a moment-matching spirit to estimate parameters of time series models such as ARMA or state-space models.
The GMWM was initially motivated by the need to estimate the parameters of complex state-space models used in various engineering applications. In short, this approach uses the quantity called Wavelet Variance (WV) in the spirit of a GMM estimator. This method is often the only feasible estimation approach that can be applied for complex models which are used in engineering and natural sciences. In particular, the GMWM is computationally efficient and, unlike most likelihood-based techniques, it can be applied to massive time dependent datasets which are becoming increasingly common. For example, one of the first applications of this method was in the field of engineering where the GMWM was used to solve "sensor calibration" problems which are of great interest in different domains such as Aerospace, Robotics or Geomatics and entails large amounts of data (typically tens of millions of observations). In this context the GMWM has been demonstrated to represent a considerable improvement compared to benchmark methods [see e.g. @stebler2014generalized for details] both in terms of statistical accuracy and computational efficiency.
Building on the generality and flexibility of the GMWM, the estimation framework was enlarged to also include robust estimators, leading to a robust version of GMWM (RGMWM), in @guerrier2020robust. Due to its computational efficiency, the GMWM is able to easily estimate complex time series and spatial models in circumstances where traditional methods have considerable computational and numerical issues, adding the robust estimation layer with only a marginal increase in computational complexity.
Below are examples of the features of the `gmwm` package.
To start, let's generate a time series from a simple model, which is a AR(1) process with measurement error (white noise):
```{r}
# Sample size
n = 10^4
# Specify model
model = AR1(phi = .98, sigma2 = .02) + WN(sigma2 = 1)
# Generate Data
Xt = gen_gts(n = n, model = model)
```
Once we have data, we can see what the wavelet variance looks like:
```{r, fig.align='center', fig.width=4, fig.height=3}
# Compute Haar WV
wv_Xt = wvar(Xt)
plot(wv_Xt)
```
```{r, eval = F}
wv_Yt = wvar(Yt)
In the second time series, we introduce a few (1%) of "extreme" (outliers):
# Copy data and add "outliers"
Yt = Xt
Yt[sample(1:n, round(0.01*n))] = rnorm(round(0.01*n), 0, 3^2)
# Plot the data
plot(wv.classical)
# Calculate robust wavelet variance
wv.robust = wvar(d, robust = TRUE, eff = 0.6)
# Compare both versions
compare_wvar(wv.classical, wv.robust)
```
Now, let's try to estimate it with specific (e.g. user supplied) and guessed (e.g. program generated) parameters.
```{r, eval = F}
## Estimation Modes ##
# Use a specific initial starting value
o.specific = gmwm_imu(AR1(phi=.98,sigma2=.05) + WN(sigma2=.95), data = d)
# Let the program guess a good starting value
o.guess = gmwm_imu(AR1()+WN(), data = d)
```
To run inference or view the parameter estimates, we do:
```{r, eval = F}
## View Model Info ##
# Standard summary
summary(o.specific)
# View with asymptotic inference
summary(o.specific, inference = T)
# View with bootstrapped inference
summary(o.specific, inference = T, bs.gof = T)
```
Alternatively, we can let the program try to figure out the best model for the data using the Wavelet Information Criteria (WIC):
```{r, eval = F}
## Model selection ##
# Separate Models - Compares 2*AR1() and AR1() + WN() under common model 2*AR1() + WN()
# Note: This function created a shared model (e.g. 2*AR1() + WN()) if not supplied to obtain the WIC.
ms.sep = rank_models(AR1()+WN(), 2*AR1(), data = d, model.type="imu")
# Nested version - Compares AR1() + WN(), AR1(), WN()
ms.nested = rank_models(AR1()+WN(), data = d, nested = TRUE, model.type = "imu")
# Bootstrapped Optimism
ms.bs = rank_models(AR1()+WN(), WN(), data = d, bootstrap = TRUE, model.type = "imu")
# See automatic selection fit
plot(ms.sep)
# View model picked:
summary(ms.sep)
```
Last, but certainly not least, we can also approximate a contaminated sample with robust methodology:
```{r, eval = F}
## Data generation ##
# Specify model
model = AR1(phi = .99, sigma2 = .01) + WN(sigma2 = 1)
# Generate Data
set.seed(213)
N = 1e3
sim.ts = gen_gts(n, model)
# Contaminate Data
cont.eps = 0.01
cont.num = sample(1:N, round(N*cont.eps))
sim.ts[cont.num,] = sim.ts[cont.num,] + rnorm(round(N*cont.eps),0,sqrt(100))
# Plot the data
plot(sim.ts)
# Classical Wavelet Variance
wv.classic = wvar(sim.ts)
# Robust Wavelet Variance
wv.robust = wvar(sim.ts, robust = TRUE, eff = 0.6)
# Plot the Classical vs. Robust WV
compare_wvar(wv.classic, wv.robust, split = FALSE)
# Run robust estimation
o = gmwm_imu(model, sim.ts, robust = TRUE, eff = 0.6)
# Robust information
summary(o)
```
## Installing the package through CRAN (Stable)
The installation process with CRAN is the simplest
```{r, eval = F}
install.packages("gmwm")
```
Installing the package this way gives you access to stable features. Furthermore, the installation itself does not require a compiler or preinstalling any dependencies. However, we are limited to updating the package on CRAN to once every month. Thus, there may be some lag between when features are developed and when they are available on this version.
## Installing the package through GitHub (Developmental)
For users who are interested in having the latest and greatest developments withing wavelets or GMWM methodology, this option is ideal. Though, there is considerably more work that a user must do to have a stable version of the package. **The setup to obtain the development version is platform dependent.**
Specifically, one **must** have a compiler installed on your system that is compatible with R.
For help on obtaining a compiler consult:
* [OS X](http://thecoatlessprofessor.com/programming/r-compiler-tools-for-rcpp-on-os-x/)
* [Windows](https://cran.r-project.org/bin/windows/Rtools/)
Depending on your operating system, further requirements exist such as:
**OS X**
Some user report the need to use X11 to suppress shared library errors. To install X11, visit [xquartz.org](http://www.xquartz.org/)
**Linux**
Both curl and libxml are required.
For **Debian** systems, enter the following in terminal:
```{r, eval = F, engine='bash'}
sudo apt-get install curl libcurl3 libcurl3-dev libxml2 libxml2-dev
```
For **RHEL** systems, enter the following in terminal:
```{r, eval = F, engine='bash'}
sudo yum install curl curl-devel libxml2 libxml2-dev
```
**All Systems**
With the system dependency taken care of, we continue on by installing the R specific package dependencies and finally the package itself by doing the following in an R session:
```{r, eval = F}
# Install dependencies
install.packages(c("RcppArmadillo","ggplot2","reshape2","devtools","knitr","rmarkdown"))
# Install the package from GitHub without Vignettes/User Guides
devtools::install_github("SMAC-Group/gmwm")
# Install the package from GitHub with Vignettes/User Guides
# Note: This will be a longer install as the vignettes must be built.
devtools::install_github("SMAC-Group/gmwm", build_vignettes = TRUE)
```
# Licensing
The license this source code is released under is the GNU AFFERO GENERAL PUBLIC LICENSE (AGPL) v3.0. In some cases, the GPL license does apply. However, in the majority of the cases, the license in effect is the GNU AFFERO GENERAL PUBLIC LICENSE (AGPL) v3.0 as the computational code is heavily dependent on Armadilllo, which use the MPL license that enables us to recast our code to use the GNU AFFERO GENERAL PUBLIC LICENSE (AGPL) v3.0. See the LICENSE file for full text. Otherwise, please consult [TLDR Legal](https://tldrlegal.com/license/gnu-affero-general-public-license-v3-(agpl-3.0)) or [GNU](https://www.gnu.org/licenses/agpl-3.0.en.html) which will provide a synopsis of the restrictions placed upon the code. Please note, this does NOT excuse you from talking about licensing with a lawyer!