9,854 reputation
12655
bio website fromthebottomoftheheap.net
location Regina, Canada
age 37
visits member for 3 years, 7 months
seen 16 hours ago

I'm Quantitative Environmental Scientist in the Institute of Environmental Change & Society, at the University of Regina, Canada. I undertake research on environmental problems, including climate change and atmospheric pollution, affecting lakes. I use lake sediments to look back in time at the history of lakes to look at what organisms are present and how the species in the lake have changed through time and how lakes evolve and respond to pollution and perturbations.

I'm also an Adjunct Professor in the Department of Biology at the University of Regina.


Apr
15
awarded  Nice Answer
Apr
10
accepted What is the autocorrelation function of a time series arising from computing a moving standard deviation?
Apr
10
revised Conceptual issues in training neural network and learning curve
typo in title
Apr
8
comment Temporal Correlation in R
@user3511094, if you care to identify yourself, feel free to contact me off list (see email address in my profile) as I have been doing these sorts of models recently.
Apr
8
comment Temporal Correlation in R
Tweddie distributions are often being used for rainfall data (they are defined for the set of positive reals but with a point mass on zero). Other approaches include a two-stage model; a binary model to model probability of rainfall and then a truncated count model for the amount of rainfall, which only kicks in if a day is predicted to have rainfall given the data. The Tweedie of a Gamma distribution could be used for the other continuous data too.
Apr
5
comment What is the autocorrelation function of a time series arising from computing a moving standard deviation?
@AlecosPapadopoulos Sorry for the confusion, the detrended series is not $MA$ - I think I mentioned this in the earlier comments. All we should assum about the series to which we are applying the moving window to generate the SD series is that it has been detrended. My reason for mentioning the $MA$ was in relation to the dependence structure for the derived SD series and it was a query - I was wondering if the dependence structure for the SD series might be similar to an $MA$ which is a weighted average of random variables. I wondered this because each SD value is a function of past values
Apr
4
comment What is the autocorrelation function of a time series arising from computing a moving standard deviation?
@AlecosPapadopoulos I would assume so; in many cases, the series in question is derived by detrending the original values through the use of a scatterplot smoother such as LO(W)ESS or a Gaussian kernel smoother. I think it is safe to assume that these residuals are correlated (not independent) Gaussian random variables, yes.
Mar
27
answered Can I think of the level of a hypothesis test as being the probability the null hypothesis is true?
Mar
27
revised How can a distribution have infinite mean and variance?
add user's comment into his question as this information gives context for the question itself.
Mar
27
comment What is the autocorrelation function of a time series arising from computing a moving standard deviation?
Having done a bit more background reading on models for the variance of a series, I wonder if it wouldn't be better all round just to fit that model than worry about the moving window bits. An (G)ARCH or stochastic volatility model seems appropriate for that at the moment, but I'm not sure how would show that variance increased with one of these models? But that's for a different Q&A. Still very much interested in any thoughts on the Q here as it is something one quite often in looking for early warning signals of impending transition in ecology.
Mar
27
comment What is the autocorrelation function of a time series arising from computing a moving standard deviation?
@Glen_b We could fit an $mathrm{MA}(q)$ to the original series, but I was more wondering if, because the original series is actually residuals after any trend has been estimated and removed, computing the mean in the moving window (in same way as I described above for the SD) would give something like an MA process and hence if there was a similar link such that the moving-SD would have ACF with similar properties to the MA process (signif correlations at lags up to $q$).
Mar
27
revised What is the autocorrelation function of a time series arising from computing a moving standard deviation?
A sentence was just plain incoherent
Mar
27
asked What is the autocorrelation function of a time series arising from computing a moving standard deviation?
Mar
26
comment Logistic regression using penalized likelihood (lasso?) in Matlab/R
@Maddy & I also misspoke (miswrote) about lasso reducing the bias. It tries to reduce variance due to estimating $\beta$s for all covariates at expense of a bit of _increased_ bias in the remaining $\beta_{\mathrm{lasso}}$. If the reduction in variance exceeds the increase in bias, then MSE will be reduced. That's what the Lasso aims to get at.
Mar
26
comment Logistic regression using penalized likelihood (lasso?) in Matlab/R
@Maddy Oops, brain fail on my part. I was thinking of coordinate descent not gradient descent, and ignore my loss function ramblings. Clearly I had gradient boosting on my mind when I wrote the comment. The idea of using L1, L2 and other penalties in GLMs is well established with fast cyclic coordinate descent algorithms for many GLM models now developed. With The lasso penalty, the idea is that some of the covariates have zero or near zero $\beta$s, i.e. the solution is sparse. If the truth really is sparse, then estimating $\beta$s for all covariates is overfitting
Mar
26
revised Logistic regression using penalized likelihood (lasso?) in Matlab/R
added 2 characters in body
Mar
26
comment Logistic regression using penalized likelihood (lasso?) in Matlab/R
@Maddy the lasso can be used in a logistic regression (IIRC you can rewrite the path algorithm as a gradient descent algorithm minimising a given loss function, which for logistic regression includes two options). It is there to reduce bias in the model estimates arising from fitting a large number of parameters. See the glmnet package for R as one implementation. I don't know how it would help in the small number of 1s case though? It doesn't deal with correlated predictors though, unlike ridge regression. The elastic net combines lasso & ridge penalties to handle sparsity & collinearity.
Mar
24
answered Logistic regression using penalized likelihood (lasso?) in Matlab/R
Mar
7
revised The difference between SVD and SVD++
edited title
Feb
25
revised How to construct a design matrix for coxph with pspline term?
added 10 characters in body