JuliaAI
Towards validating tutorials render correctly on the website
I expected to fix the plots for the files that were broken earlier hence the branch name but after running them locally now (and checking the corresponding PRs) seems like @ablaom has fixed them as well. Thank you for that.
In the commit above I add missing Project.toml and Manifest.toml files for the new credit fraud tutorial.
In the commit above non-code in seven tutorials was modified to let serving pass. This includes exactly two things:
- Some
@@were altered (removed) when these tutorials were fixed. I ran the script again after removing them all and their Franklin errors were gone. @ablaom I think we may need to clarify in the README that contributors should ignore them when modifying existing tutorials (or I can easily modify the script to handle fixing them) and one would have to rerun after changing tutorials. - Removing special characters from comments inside functions before a return. An ambiguous error due to Franklin would arise otherwise.
Note that the fact that serving passed doesn't imply that everything will render correctly as code errors are treated as warnings. This is what I will be looking into now.
Data Basics Equivalence Check
Check is done by comparing old and new website side by side for the specific tutorial.
| Tutorial Name | Check Results |
|---|---|
| D0-categorical | Success |
| D0-dataframe | Success |
| D0-loading | Success |
| D0-processing | ~~Need to fix Franklin Warning Failed to precompile Plots.~~ Success; also had to addCSV. |
| D0-scitype | Success |
~~A note about the error above is that (most) plots still show successfully but the warning shows as an error in the notebook. Edit: NVM. It shows old plots of the same name...~~
Getting Started Equivalence Check
| Tutorial Name | Check Results |
|---|---|
| A-model-choice | Success |
| A-fit-predict | Success |
| A-model-tuning | Success but minor inconsistency due to RNG |
| A-ensemble | Success but minor inconsistency due to RNG |
| A-ensemble-2 | Success but very minor inconsistency due to RNG |
| A-composing-models | Success but actually modified |
| A-ensemble-3 | Success but actually (minorly )modified |
| Stacking | Success but actually modified |
Introduction to Statistical Learning Equivalence Check
| Tutorial Name | Check Results |
|---|---|
| Lab 2 | Success with minor RNG issue |
| Lab 3 | Success |
| Lab 4 | Success after typo fix |
| Lab 5 | Success |
| Lab 6b | Success with likely RNG issues |
| Lab 8 | Success |
| Lab 9 | Success |
| Lab 10 | Success |
| Tutorial Name | Check Results |
|---|---|
| EX-telco | Success but actually modified |
| EX-GLM | Success |
| EX-airfoil | Success |
| EX-boston-flux | Success |
| EX-breastcancer | Success |
| EX-horse | Success except for slightly inconsistent conclusions in A baseline model section |
| EX-housekingcounty | Success |
| EX-powergen | Success |
| EX-wine | Success ~except for a very minor conclusion in Getting a Baseline section~ |
| EX-AMES | Success ~but may be misunderstanding something as an inconsistency~ |
| EX-crabs-xgb | Success ~with a plot-conclusion inconsistency in XGBoost Machine as an exception~ |
| ⛔ EX-boston-lgbm | Unable to run full tutorial due to NULL library handle |
Summary of Checks:
- All plots render as expected except for one instance as reported in the last table
- All outputs are okay except for some instances where the conclusions didn't quite align with them as reported in the last table
- I was unable to
serve()LightGBM tutorial, likely due to platform issues, as reported in the last table
I have also added a commit where I have increased the font sizes for some tutorials. Some tutorials had ordinarily-sized and legible text plots so I didn't touch them. Otherwise, I increased font scaling then viewed the tutorial again. I don't always scale with the same number because when plot(machine) is present a smaller font is preferred to prevent text from clashing with the axes.
Take note that plots may appear larger than expect while serving. I wasn't able to solve this by changing the figure size with Plots.jl but was able to solve it by limiting image sizes via CSS (but I didn't push any style changes).
Will say that for me, Franklin's serve() with an empty directory doesn't behave the same as serve() followed by serve(eval_all=true) which is surprising (and took me time to figure out).
Think I am now done with validating the tutorials and plots for all tutorials except for LightGBM.
I will now move on to other PRs related to the navigation bar items or styling.
@EssamWisam Are you able to resolve the conflicts here? Or is this PR now redundant?
@EssamWisam Are you able to resolve the conflicts here? Or is this PR now redundant?
This is the most important PR on the table xD. I think solving the conflict should be easy. It's just because in the other PR we merged already I moved A-ensembles-3 to advanced.
I found some broken links here (first two):
But actually as these resources are now quite old, let's just remove them (keeping the preceding link to Telco). I'm probably not going to update them soon.
The link to external resources points to a non-existent file. You can find what was planned for this here.
Let's remove "MLJTutorials" from this list, and all other references to it, as it is quite old.
A small annoyance is that switching tabs on the landing page is not permanent, in the sense that going back after following a link always returns you to the "New to Julia?" tab. If this is a lot of work to fix, however, let's just create an issue to flag it for later.
The first link here is broken:
And "top of this page" no longer makes sense. Maybe just "click here".
A bunch of links from the tabs to intro to stats learning are broken. I think there is a spelling mistake (my bad):
("learning" not ".earning")
@ablaom The commits above address all the commends you mentioned. I tried to be cautious while testing so I can merge this directly.