Bug fix for moving average tracking #48
Conversation
nathanielpritchard
commented
May 12, 2024
nathanielpritchard
commented
- Added more functionality to the moving average tracking to allow for use of a short width and long width moving average, which I thought worked but upon a closer look did not work.
| res::Float64 = log.dist_info.scaling * (eltype(samp[1]) <: Int64 || size(samp[1],2) != 1 ? | ||
| log.resid_norm(samp[3])^2 : log.resid_norm(dot(samp[1], x) - samp[2])^2) |
There was a problem hiding this comment.
Why is there so much white space added in?
There was a problem hiding this comment.
At the time I thought it was more clear to have the if else statement within the parentheses aligned. I know see that is not the case and I have removed it.
| # Check if MA is in lambda1 or lambda2 regime | ||
| if ma_info.flag | ||
| update_ma!(log, res, ma_info.lambda2, iter) | ||
| else | ||
| #Check if we can switch between lambda1 and lambda2 regime | ||
| if res < ma_info.res_window[ma_info.idx] | ||
| if iter == 0 || res <= ma_info.res_window[ma_info.idx] |
There was a problem hiding this comment.
This line needs more explanation
There was a problem hiding this comment.
I have clarified that this lean is checking for a monotonic decrease in the sketched residuals
|
|
||
| elseif ma_info.lambda == ma_info.lambda1 && !ma_info.flag | ||
| diff = ma_info.idx - ma_info.lambda | ||
| # Determine start point for first loop | ||
| startp1 = diff < 0 ? 1 : (diff + 1) | ||
|
|
||
| # Determine start and endpoints for second loop | ||
| startp2 = diff < 0 ? ma_info.lambda2 + diff + 1 : 2 | ||
| endp2 = diff < 0 ? ma_info.lambda2 : 1 | ||
| # Compute the moving average two loop setup required when lambda < lambda2 | ||
| for i in startp1:ma_info.idx | ||
| accum += ma_info.res_window[i] | ||
| accum2 += ma_info.res_window[i]^2 | ||
| end | ||
|
|
||
| for i in startp2:endp2 | ||
| accum += ma_info.res_window[i] | ||
| accum2 += ma_info.res_window[i]^2 | ||
| end | ||
|
|
||
| #Update the log variable with the information for this update | ||
| if mod(iter, log.collection_rate) == 0 || iter == 0 | ||
| push!(log.lambda_hist, ma_info.lambda) | ||
| push!(log.resid_hist, accum / ma_info.lambda) | ||
| push!(log.iota_hist, accum2 / ma_info.lambda) | ||
| end |
There was a problem hiding this comment.
This conditional statement needs more commentary. When is this accessed?
There was a problem hiding this comment.
I see now that this is really unclear, I have added comments that I think clarify what is going on
| estimate. By default, this is set to `1`. | ||
| - `sigma2::Union{Float64, Nothing}`, the variance parameter in the sub-Exponential family. | ||
| If not specified by the user, a value is selected from a table based on the `sampler`. | ||
| If the `sampler` is not in the table, then ?. |
There was a problem hiding this comment.
Replace the ? with default behavior
| If the `sampler` is not in the table, then ?. | ||
| - `omega::Union{Float64, Nothing}`, the exponential distrbution parameter. If not specified | ||
| by the user, a value is selected from a table based on the `sampler`. | ||
| If the `sampler` is not in the table, then ?. |
There was a problem hiding this comment.
Replace the ? with default behavior
