Test adding ARI to a New Model Run #245
Conversation
Damonamajor
requested review from
dfsnow,
wrridgeway and
jeancochrane
as code owners
dvc.yaml
Outdated
There was a problem hiding this comment.
nitpick (blocking): Most of the changes here just move stuff around. Let's revert the changes to this file, since they don't actually impact the outcome.
params.yaml
Outdated
| run_id: "2024-03-17-stupefied-maya" | ||
| run_id: "2024-06-02-test-damon" |
There was a problem hiding this comment.
issue (blocking): The export.run_id key is really only used for changing the outputs of the export pipeline stage. If you want to specify a run_id, use a parameter in the YAML header of your Quarto doc.
pipeline/00-ingest.R
Outdated
There was a problem hiding this comment.
issue (blocking): It's fine to edit this when messing around, but we shouldn't include changes to this file with the merge request. See my earlier structure note for a workaround.
ARI-model-test.qmd
Outdated
There was a problem hiding this comment.
issue (blocking): Before I review the content of this report, let's get the structure worked out. Let's make a new root-level directory (analyses) where one-off analyses related to the model can live. Then we'd have:
reports/for things that are generated with every runanalyses/for things that are created one time
In order to make the analysis reproducible, you need to load data from sources accessible to others. That would mean:
- For files in
input/, push to the S3 DVC cache, then useread_parquet()to load the file you pushed directly from S3 - For files in
output/, finish a run and push it to S3. Then load the files directly from S3, again usingread_parquet()
Note that you can re-use the reports/_setup.qmd file here, as it will load the input data from DVC (assuming you pushed) and most relevant output data for a given run.
There was a problem hiding this comment.
These should all be fixed now. I pushed to DVC, let me know if there are any other steps that should be taken.
|
| library(sf) | ||
| library(dplyr) | ||
| library(ggplot2) | ||
| library(noctua) | ||
| library(arrow) | ||
| library(kableExtra) | ||
| library(spdep) | ||
| library(leaflet) |
There was a problem hiding this comment.
nitpick: Alphabetically order your dependencies.
| title: "Testing the Impact of the Affordability Risk Index on the Model" | ||
| format: html | ||
| --- | ||
| ```{r, include = FALSE} |
There was a problem hiding this comment.
nitpick: Run styler::style_file() on this and lintr::lint(). There are some minor formatting issues in here.
| con <- dbConnect(noctua::athena()) | ||
| noctua::noctua_options(cache = 10) | ||
|
|
||
| input_data <- read_parquet("input/assessment_data.parquet") %>% |
There was a problem hiding this comment.
issue (blocking): This line is the main point of difference between this and the other reports. The other reports look at the current data, while this report should look at a snapshot of data that includes the variable of interest. In other words, this should ingest input data from DVC and output data from S3 so that we have a fixed, deterministic report for each feature.
| x <- round(x, 2) | ||
| print(paste("Correlation =", x, "when removing 0s")) | ||
| ``` | ||
| ## Creating a map to see if there are spatial disparities in the inter-tract SHAP values |
There was a problem hiding this comment.
nitpick: Use consistent title case.
Co-authored-by: Dan Snow <31494343+dfsnow@users.noreply.github.com>
Co-authored-by: Dan Snow <31494343+dfsnow@users.noreply.github.com>
Co-authored-by: Dan Snow <31494343+dfsnow@users.noreply.github.com>
Co-authored-by: Dan Snow <31494343+dfsnow@users.noreply.github.com>
Co-authored-by: Dan Snow <31494343+dfsnow@users.noreply.github.com>
Co-authored-by: Dan Snow <31494343+dfsnow@users.noreply.github.com>
Damonamajor
changed the title
…vm into Test_New_Model_Run
Attached is a markdown file, as well as changes to the model pipeline to allow a test of the Affordability Risk Index. Mostly for Dan, but let me know where it should be housed. It shouldn't be merged, but seemed like it would be easiest to review / look at in this folder.