Meeting June 13, 2019 - Githubissues

Sayani07 / gsoc2019

Google Summer of Code 2019 application template

0 stars 0 forks source link

Meeting June 13, 2019 #5

Open Sayani07 opened 5 years ago

Sayani07 commented 5 years ago

A walk through of the package and associated functions
Need examples for each function to demonstrate it's usage
A matrix of time granularities showing clashes and granularities will be helpful
bikedata analysis might be useful in this case.

For example, it would be very interesting for users to know when stations are empty (no point in looking there) or full (can’t leave your bike there). The regulations for a couple of cities indicates that the trips that are recorded may not be the trips that were intended, either because you have to go further to find a bike or because you have to leave your bike at a different station than you planned (or both). It would be interesting to analyse the numbers of bikes on trips from a particular station and the numbers of bikes on trips to the same station and look at how those series develop over time.

Sayani07 commented 5 years ago

Included a function comptbl_set in gravitas which gives the following matrix as output .

comp_tbl(.data, "hour", "week")

granularities hour_day hour_week day_week

1 hour_day FALSE FALSE TRUE 2 hour_week FALSE FALSE FALSE 3 day_week TRUE FALSE FALSE This is a symmetric matrix. Diagonal elements represents taking same granularity across x-axis and facets which is a clash, by definition. If the pair is harmony then the matrix provides TRUE in the corresponding cell. Also, the order of the granularities are by length of time and not alphabetically. -`comp_tbl(.data, "hour", "day")` Error in comp_tbl(.data, "hour", "day") : Only one granularity hour_day can be formed. Function requires checking compatibility for bivariate granularities - This error handling added

Sayani07 commented 5 years ago

Suggestions from Antony to incorporate -

The ordered matrix is better, although visually not great. You could just use blank for FALSE and a symbol for true. A matrix may not be the best representation. Although you treat pairs equally (e.g., hour-day and day-week), they do have an ordering in the sense that hour-day comes first. Would a directed graph be worth trying?

The pairing day-fortnight and week-month puzzles me. Would it not clash? And if it does not, when might it be useful?

Sayani07 commented 5 years ago

To follow current state of bikedata analysis - data_prep01.R and analysis02.Rmd