Open codetalker7 opened 2 years ago
@codetalker7, any update on this issue?
@sourish-cmi no update as such. I had just copied the old print messages (i.e from the old code) and put them in this file.
For Bayesian models, we just use the show
method on the chain
that Turing.jl returns.
For the frequentist models, the old code contained a few print statements; it'll be great if we can discuss proper printing messages for all these.
@codetalker7, In the frequentist method, I think the output would be
Model Class: Linear Regression
Likelihood Mode: Gauss
Link Function: Identity
Computing Method: Optimization
───────────────────────────────────────────────────────────────────────────────────────
Coef. Std. Error t Pr(>|t|) Lower 95% Upper 95%
───────────────────────────────────────────────────────────────────────────────────────
(Intercept) 5.7696e5 10709.5 53.87 <1e-99 5.55934e5 5.97985e5
Area 453107.0 18295.7 24.77 <1e-97 4.17188e5 4.89026e5
Room -1.18099e5 17251.0 -6.85 <1e-10 -1.51967e5 -84230.5
Area & Room 50745.1 4369.19 11.61 <1e-27 42167.2 59322.9
───────────────────────────────────────────────────────────────────────────────────────
R-squared: 0.749320
I would like to see a similar table for Bayesian regression - something like this
Model Class: Linear Regression
Likelihood Model: Gauss
Prior Model: Ridge
Link Function: Identity
Computing Method: HMC
───────────────────────────────────────────────────────────────────────────────────────
Coef. Std. Dev Lower 95% Upper 95%
───────────────────────────────────────────────────────────────────────────────────────
(Intercept) 5.7696e5 10709.5 5.55934e5 5.97985e5
Area 453107.0 18295.7 4.17188e5 4.89026e5
Room -1.18099e5 17251.0 -1.51967e5 -84230.5
Area & Room 50745.1 4369.19 42167.2 59322.9
───────────────────────────────────────────────────────────────────────────────────────
I see. @sourish-cmi if possible, can you provide the Model Class
, Lilkelihood Model
, Link Function
and Computing Method
for all frequentist models?
Or rather, are the current ones in src/print.jl
correct (these are passed as string arguments to the details
function in that file).
For the Bayesian models, we already have two tables printed out: Summary Statistics
and Quantiles
, which I believe contain the given information. Apart from this, we can add the Model Class
, Likelihood Model
, Prior Model
, Link Function
and Computing Method
for Bayesian models too.
@codetalker7 currently, the frequentist method prints Model Class
, Likelihood Model
, Link Function
and Computing Method
. So we don't have to do anything.
You are correct. The Bayesian models have printed out all the necessary information. But if the print
function can print the same information in the same format as the frequentist method, that will bring nice conformity in the presentation of the results.
The file for this issue is src/print.jl. We have extended the
Base.show
method to print details about models when they are trained; currently these details are not filled in appropriately, and might need some improvements.