| 12345678910111213141516171819202122232425262728293031323334353637 |
- Changes to Chronux from previous version
- 1) Better comments particularly in the coherency calculations - We now explain that the
- coherency routines return the standard deviation of phi in the variable phierr. However,
- in order to ensure that the calling sequence remains the same when Jackknife calculations
- are put in, the first dimension of phierr can be 1 or 2 corresponding to upper and lower
- confidence bands. Thus phierr(1,...)=phierr(2,...)=standard deviation of phi. For p=0.05
- band one should then take phi+/- phierr(1,...). When Jackknife calculations are introduced,
- we will eventually have phierr(1,...) not equal to phierr(2,...). In that case the correct
- confidence intervals will be given by phi-phierr(1,...) and phi+phierr(2,...).
- 2) There was a problem with previous calculations of the derivatives. In particular, an
- approximate formula for the derivatives is given by X' A X where X=(x(1),x(2),...,x(k)) are
- the k fourier transforms (each a function of frequency), X' is the adjoint of X and A is a
- bidiagonal matrix of the form A(i,j)=sqrt(i) for j=i+1; A(i,j)=sqrt(j) for j=i-1, and A(i,j)=0,
- otherwise. The previous formula assumed A was bidiagonal but with the off diagonal terms all
- being 1.
- 3) We have introduced routines for doing the non-stationarity test a la Thomson (2000). These routines
- are currently available only for continuous data.
- 4) We have now put in routines doing the two group comparison of coherences and spectra
- (Bokil et al, to be submitted, 2006). These are applicable to point as well as continuous processes
- and allow you to compute a test statistic \Delta z, and its Jackknife variance. The test statistic
- is distributed as N(0,1) under the null hypothesis of equal population coherences or spectra
- when the underlying data are non-Gaussian. The routine also returns a variable that contains information
- about the frequencies at which the null can be rejected. In addition,large deviations of the Jackknife
- variance from 1 occurs at frequencies where the data shows non-gaussian fluctuations. The test should
- be treated with considerable caution at these frequencies. These are the directory chronux/statistical_tests
- 5) New routines have been written to compute the cross-spectral matrix, its singular values and leading
- singular vectors, as well as the Wiener entropy for point process data. The routine for continuous valued
- data is now names CrossSpecMatc.m with the old routine CrossSpecMat.m avalable in the directory chronux/old.
- 6) Fixed an error in binspikes.m that caused some problems with the time
- window specified was smaller than the maximum duration of the spikes.
|
