robjhyndman / anomalous Goto Github PK

I'm curious as to how pcaPP::PCAproj differs from "regular" principle components analysis. Why'd you choose to use it in this package?

I am using R 3.4.0 on Debian 8.1. When I tried to install the package I have this error:

ERROR: dependency 'mvtnorm' is not available for package 'pcaPP'
* removing '/usr/local/lib/R/site-library/pcaPP'
Installation failed: Command failed (1)
'/usr/lib/R/bin/R' --no-site-file --no-environ --no-save --no-restore --quiet  \
  CMD INSTALL  \
  '/tmp/RtmpcMckxO/devtools21581c951b67/robjhyndman-anomalous-acm-79d50aa'  \
  --library='/usr/local/lib/R/site-library' --install-tests

ERROR: dependencies 'ForeCA', 'alphahull', 'pcaPP', 'hdrcde' are not available for package 'anomalousACM'
* removing '/usr/local/lib/R/site-library/anomalousACM'
Installation failed: Command failed (1)

Any suggestion?

Thanks a lot

Hi Rob,
first of all thank you for your great packages. Do you think it's possible to make a release and tag it? As suggested here? I'm trying to build a conda package for this.

Can you guide me in lay programmer's term on how to use the APIs(if only there's a provision for it).
I've the data for each day of the year and whenever the new data point comes in I want to check whether it's anomalous to the time series or not.
Thanks.

I'm getting the following error when I try to run the example.
y<-tsmeasures(dat0)

Error in (function (env, objName) :
could not find function "object.size"

Hi,

Very nice package, thank you.

Though, I feel like this package could use another small feature. The n parameter in the anomaly function basically returns n most "anomalous" time-series, while we can basically return them all (if n == nrow(x)). In other words, we don't have some significance measure that will tell us how significantly anomalous those series are and we can't specify our desired significance threshold (in other words, we don't know how to specify n correctly).

I've took a look in the anomaly function source code and saw that this is already being calculated using the hdrcde::hdr.2d function and it could pretty easily be incorporated into the anomaly function.

I don't know if this is really makes sense, so I haven't made it a PR. Instead, I've made a GIST that adds a parameter called sig. If speciffied, it will return/plot the time series that are sig <= hdrinfo$falpha . I've also tried to make sure that it is backward compatible and won't brake any existing code.

Please advise,
Thanks again,
David