astrogilda / tsbootstrap

tsbootstrap: generate bootstrapped time series samples in Python

https://tsbootstrap.readthedocs.io/en/latest/

MIT License

65 stars 5 forks source link

125 update roadmap #126

Closed astrogilda closed 2 months ago

astrogilda commented 3 months ago

Pull Request Template

Description

Updated roadmap in README.md. closes #125 Expanded roadmap at #144

Type of change

Please delete options that are not relevant.

[ x] This change requires a documentation update

Checklist:

[ x] My code follows the style guidelines of this project
[ x] I have performed a self-review of my own code
[ x] I have made corresponding changes to the documentation
[x ] My changes generate no new warnings

Add All Contributors Command

Remember to acknowledge your contributions, replace contribution_type with your contribution (code, doc, etc.):

@all-contributors please add @<your-username> for <contribution_type>

astrogilda commented 2 months ago

It seems like this PR accidentally contains commits from multiple PR. Accidental push?

Can you kindly untangle this?

Just rebased against main, should be g2g!

fkiraly commented 2 months ago

Just rebased against main, should be g2g!

My recommendation is merging from main. I know there are differring schools here, I just feel the risk for unintended reverts (like the above?) is less if you do things quickly.

fkiraly commented 2 months ago

Some random ideas for roadmap:

integration

using arbitrary sktime forecaster in forecaster based bootstraps
using tsbootstrap bootstraps to create a distribution or sampler-like object (skpro or sktime proba distribution over time series)

boilerplate/sugar

fractions for block lengths etc
ability to pass pd.DataFrame. "core" behaviour is np.ndarray, but if pd.DataFrame is passed, produces the output to have same index and columns

refactor

there is some legacy entanglement of classes. These should be refactored to be "flatter".
the block bootstraps look like they should be composites

API extension

update/stream
panels and hierarchical data
exogenous data (API supports but there is no logic except for forecasters)
fittable or pretrained

adjacent areas

time series augmentation
full proba models that can be sampled from
time series data generation

astrogilda commented 2 months ago

Hmm this is after rebasing against main. Maybe let's focus on the readme.md file for now given the time constraints?

On Mon, Apr 22, 2024, 5:34 PM Franz Király @.***> wrote:

@.**** requested changes on this pull request.

still something weird going on with all files being changed.

— Reply to this email directly, view it on GitHub https://github.com/astrogilda/tsbootstrap/pull/126#pullrequestreview-2015841406, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFTOOHQXUOI5QDLEAABPU63Y6V65BAVCNFSM6AAAAABF2ZTWXOVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDAMJVHA2DCNBQGY . You are receiving this because you authored the thread.Message ID: @.***>

astrogilda commented 2 months ago

it's linked in the updated readme.py

fkiraly commented 2 months ago

you can follow a hash if it is in a markdown doc in the repo? I thought that worked only in issues and PR?

astrogilda commented 2 months ago

Just tried, I guess you can't. I'll put it here.

fkiraly commented 2 months ago

you could just add an ordinary link, like [name](URL)

astrogilda commented 2 months ago

sure. let's please approve it after this.