Skip to content

Home

Jump to bottom
mj-xmr edited this page Feb 9, 2022 · 61 revisions

Welcome to the tsqsim wiki!

Development plan

Major

  • documentation, documentation, documentation.
  • QT app upgrade to QT6 and QCustomPlot 2
  • QT app needs to have all its lost features (due to decoupling) restored
  • confidence bands (1st request from Rucknium)
  • weekly discrete time steps (2nd request from Rucknium). Currently the largest time steps are days. Adding monthly won't hurt either.
  • finalize the diff transformation. The reconstruction of predictions is skewed.
  • walk forward validation needs better UI and more consistency regarding data holding (adding data/records instead of replacing them for each validation window)
  • plugin interface for Predictors
  • plugin interface for Loss Functions
  • plugin interface for Optimization Goals (todo: prediction score vs. truth, alongside the already coded pred. score vs. baseline)
  • self defined transforms and scripts
  • protect against look-ahead bias (series transformations to achieve stationarity, as well as individual vector transformations, to ease the writing of optimized versions of predictors)
  • time scale manipulation
  • seasonal corrections (before the time scale manipulation!)
  • generate alternative scenarios
  • networked processing

Done

  • protect against look-ahead bias: done for IPredictor. Needs work for XForm.
  • prediction: baseline and advanced
  • optimization methods. Goal = prediction score vs. baseline prediction
  • walk forward validation
  • R integration
  • ACF, PACF & Seasonal Decomposition plots in Python

Minor

  • distribution of solutions for Grid Search, just like Monte Carlo.
  • plot how input parameters change the output variable
  • HTML reports
  • better test coverage
  • debugging methods (plot each transformation individually)
  • QT app still hides weekend data
  • speed up serialization
  • separate simulator for stationarity optimizations
  • full CLI. Currently using WX App for all configuration
  • the seasonal decomposition plot is definitely reversed. How about the other plots (ACF/PACF)?

Done minor

  • select between close / open / high / low candle data source (instead of only close)
  • Python plots (robust) as an alternative to the QT app.
  • plot the distribution of the results obtained by the MonteCarlo method
  • allow various input formats and guess which one we're dealing with

Known bugs

  • the "diff" transformation distorts the predictions (some progress was made)

Clone this wiki locally