[ENH] Add script for making descriptive bar plots #18
Conversation
There was a problem hiding this comment.
This is looking great, thanks @michellewang!
I left some suggestions to simplify the code a bit; feel free to take what makes sense to you.
In terms of layout, I'm sure Eric will have his own opinions (so we may want to wait to implement some of the following until Friday), but some ideas from me:
- Now that I'm looking at the bar plots, I'm wondering if it's more visually intuitive to have the y-axis be %, and show the actual counts on hover instead. Wdyt?
- If we do want to keep counts on the y-axis, might be worth having a y-axis title/titles that say "count (1000s)" or similar, and just make the axis tick labels integers, if that makes sense / is not too hard
- For the demographic variables w/ longer categories (political party, employment status, household origin), maybe we can wrap the x-axis tick labels so they don't extend as far. Here's a function I used to do this in the digest dashboard, for reference (there might be easier ways). If that seems like too much hassle however, we could just add more vertical space between plots so there's no overlap. totally up to you
- Since q2 is a bit unique but still has long category labels, maybe exceptionally for this plot, we could switch the orientation of x and y such that the likert responses are on the y, and the counts/proportions along the x. Perhaps that could help to a) distinguish it from the other demo variables 2) make it more readable (?)
- In general, I think a slightly bigger gap b/w subplots on the same row may be nice (the bars themselves can be fairly thin), for more visual whitespace?
- Minor, but I prefer a white background to the default seaborn (?) purplish grid background for cleaner-looking plots. Maybe there is a simpler white theme we can use?
If you strongly agree with any of the above, maybe you can implement them before merging. For the rest, we can revisit on Friday :)
Also, if you find any of the customization challenging using plotly graph_objects, I suggest checking out plotly express, which is more high-level but should have a simpler syntax (this is what I used for the dashboard)
| COL_DEMOGRAPHIC_VARIABLE = "demographic_variable" | ||
| COL_CATEGORY = "category" | ||
| COL_N = "n" | ||
| COL_PERCENTAGE = "percentage" |
There was a problem hiding this comment.
These column names can be expected to always be the same in the TSV. For simplicity/code readability, what do you think about getting rid of these global COL_ vars and just referring to the column name directly in the functions below?
| return bar_plot | ||
|
|
||
|
|
||
| def make_impact_plot_traces(data: pd.DataFrame): |
There was a problem hiding this comment.
| def make_impact_plot_traces(data: pd.DataFrame): | |
| def make_impact_plot_traces(df: pd.DataFrame): |
for consistency, since you have df as the input param for the other functions
| for category in ["No", "Yes"]: | ||
| data_category = data_impact.loc[data_impact[COL_CATEGORY] == category] | ||
| traces.append( | ||
| go.Bar( |
There was a problem hiding this comment.
For the grouped bar plot especially, you might find it easier to use plotly express instead of the more low-level graph_objects. You should be able to achieve the same goal in much fewer lines (and it may be more straightforward to customize as well). Up to you, but check out https://plotly.com/python/graph-objects/#comparing-graph-objects-and-plotly-express for an example!
| return fig | ||
|
|
||
|
|
||
| if __name__ == "__main__": |
There was a problem hiding this comment.
This is just for testing things out, right? If so, I'd say we comment this out before merging
|
Hey @Remi-Gau, pinging you here in case you have more feedback on the sample descriptive plots :) These will live in a drawer component that can be toggled visible or not in the app, see GIF preview in: |
|
Merging since I won't have time to change this before the meeting tomorrow |
Closes #5.
Feedback welcome, especially about the layout since I don't think it looks great at the moment and we will have to see how it integrates into the dashboard. Also this is using the default colours but we probably want to customize them
Example (whole sample):