Code

Time#37

Code

Merged

AlxdrPolyakov merged 50 commits intomaintransferwise/wise-pizza:mainfrom timetransferwise/wise-pizza:timeCopy head branch name to clipboardFeb 6, 2024

+147,771-157

Commits

Commits on Nov 6, 2023

• First mild code refactors towards time series handling

  EgorKraevTransferwise

  committed
  d96c39c

• Factor heuristic preselection of candidates into a separate class. All tests pass.

  EgorKraevTransferwise

  committed
  05bfd57

Commits on Nov 7, 2023

• More small steps towards time series. All tests pass.

  EgorKraevTransferwise

  committed
  f285f7f

Commits on Nov 10, 2023

• More small steps towards time series. All tests pass.

  EgorKraevTransferwise

  committed
  fd2b06d

• Make heuristic pre-selection incremental. All tests pass.

  EgorKraevTransferwise

  committed
  67b57c7

• First rough cut of time series; its test fails (commented out), all other tests still pass.

  EgorKraevTransferwise

  committed
  2d911fc

• Time series regression runs without exceptions, but detected segments look wrong. All other tests pass.

  EgorKraevTransferwise

  committed
  316c61e

Commits on Dec 1, 2023

• First working-ish version of time series-aware Wise Pizza.

  Show description for 22c4563
  EgorKraevTransferwise

  committed
  22c4563

Commits on Dec 6, 2023

• Update __init__.py

  AlxdrPolyakov

  authored
  a9fc54f

Commits on Dec 15, 2023

• Merge branch 'main' of https://github.com/transferwise/wise-pizza into time

  Show description for 5772636
  EgorKraevTransferwise

  committed
  5772636

• Post-merge commit

  EgorKraevTransferwise

  committed
  0661e3d

Commits on Dec 18, 2023

• First OK-ish cut of TS graph, not great so far

  EgorKraevTransferwise

  committed
  27146b5

• Merge branch 'time' of https://github.com/transferwise/wise-pizza into time

  Show description for c5a33db
  EgorKraevTransferwise

  committed
  c5a33db

Commits on Dec 22, 2023

• Graph starts to look better, basis still very primitive

  EgorKraevTransferwise

  committed
  249184e

• A pretty decent cut with the OMP solver

  EgorKraevTransferwise

  committed
  2f8de41

Commits on Jan 5, 2024

• An OK cut of plotting for TS; plotting for active customers still TBD

  EgorKraevTransferwise

  committed
  a19a599

Commits on Jan 8, 2024

• Much nicer plotting

  EgorKraevTransferwise

  committed
  22074e2

• Simple fixes to address memory issues for TS; not enough:)

  EgorKraevTransferwise

  committed
  f60db71

• Refactor plotting prior to doing joint fit for TS weights and totals.

  EgorKraevTransferwise

  committed
  b709785

• more refactor

  EgorKraevTransferwise

  committed
  602cacb

Commits on Jan 9, 2024

• Nice joint plotting of weights, avgs and totals

  EgorKraevTransferwise

  committed
  d755e6a

• Legend fix

  EgorKraevTransferwise

  committed
  5cc9abd

• A failed attempt to use fitted weights for volume forecasting

  EgorKraevTransferwise

  committed
  7c69690

• Refactor to limit memory consumption for TS

  EgorKraevTransferwise

  committed
  6fc6512

• Optimized memory consumption for higher depths

  EgorKraevTransferwise

  committed
  1eec9fd

Commits on Jan 11, 2024

• Fixes to pass tests

  EgorKraevTransferwise

  committed
  a3112de

Commits on Jan 12, 2024

• Comment out numba usage, didn't seem to help

  EgorKraevTransferwise

  committed
  5ba96d0

Commits on Jan 16, 2024

• Refactor TS plotting in preparation for log fits

  EgorKraevTransferwise

  committed
  f2acc61

Commits on Jan 17, 2024

• Kinda working cut of fitting weights in rescaled space.

  Show description for f6f1f46
  EgorKraevTransferwise

  committed
  f6f1f46

• Factor transforms into separate class, segment contributions still broken

  EgorKraevTransferwise

  committed
  5ff4b55

• Fitting in transformed space appears to actually work

  EgorKraevTransferwise

  committed
  fcd84b9

Commits on Jan 18, 2024

• A shot at log space fitting for totals too, broken so far

  EgorKraevTransferwise

  committed
  e300846

Commits on Jan 19, 2024

• Combine plots for segments that are identical except for time profile

  EgorKraevTransferwise

  committed
  5ace523

Commits on Jan 22, 2024

• Consistent fitting in log space almost works, only green segment lines look broken

  EgorKraevTransferwise

  committed
  37a81fe

• Move segment impact to right y axis

  EgorKraevTransferwise

  committed
  65fdec8

Commits on Jan 29, 2024

• Pre-forecasting snapshot; fitting both weights and avgs works, fitting only weights fails mysteriously

  EgorKraevTransferwise

  committed
  43b06b0

• Add a fit_sizes parameter, fitting of growth data works both with it on and off.

  EgorKraevTransferwise

  committed
  f645911

• Fix index type of extended basis

  EgorKraevTransferwise

  committed
  a014ac0

• predict() runs without exceptions

  EgorKraevTransferwise

  committed
  e1290fe

• Minor refactors on the way to plotting predictions. Fits both with and without fitting weights run fine.

  EgorKraevTransferwise

  committed
  31281af

Commits on Jan 30, 2024

• Snapshot of prediction plotting. Plot displays but is garbage.

  EgorKraevTransferwise

  committed
  e1135d7

• Plotting approaches reasonable, only a bit of a jump in the transition

  EgorKraevTransferwise

  committed
  bed7748

• Pretty decent cut of forecasting w/o weights

  EgorKraevTransferwise

  committed
  287d0d7

• First cut of blending log transform with forecasting. Plots still very suspect.

  EgorKraevTransferwise

  committed
  46955fb

Commits on Jan 31, 2024

• Tidy up the plotting data interface, fix nan scaling bug. Predict with identity transform works.

  EgorKraevTransferwise

  committed
  625c6a6

• Added weights TS to avg-only TS fit. Usage of external weights still broken.

  EgorKraevTransferwise

  committed
  38742e6

• External weights seem to work OK now, plots pretty.

  EgorKraevTransferwise

  committed
  38a7c33

Commits on Feb 1, 2024

• Forecasting with given future weights definitely works

  EgorKraevTransferwise

  committed
  c271b58

Commits on Feb 2, 2024

• added plot_is_static, colors, timeseries notebook

  

  Alexander Polyakov

  committed
  70b0b97

• added scale for impact line

  

  Alexander Polyakov

  committed
  e2cc8c2