Skip to content

Latest commit

 

History

History
53 lines (30 loc) · 2.03 KB

NOTES.md

File metadata and controls

53 lines (30 loc) · 2.03 KB

In order of Priority: IDEAS and TODOS

In order of priority:

High priority

  • Set legend false on algo ranking bar chart

  • Do target sweep with linear regression only or 1 or 2 more good algorithms.

Medium priority

  • Implement Las Vegas initialization options, in particular one with a full set of oscillators.

  • Plot Improvement: custom legend for mask with multiple members in a group. Such that a group shares a color and one entry in the legend (individual algorithm names shouldn't be visible).

  • Plot Improvement: save mean, mode, stddev to frequency diversity plot title.

  • Add magpie to target sweep.

  • Plot improvement: Reuse algorithm bar chart with z-ops sweep.

  • Python: increase sampling rate to avoid aliasing. Verify with individual oscillator plots.

  • Map R to Freq in analysis for Hybrid signal generator

  • Plot Improvement: For oscillators vs. rmse. Show up to 1k oscillators, but draw a dotted line into the plot at around 300 oscillators to indicate what's technologically feasible.

  • Reproducibility issue: Maffezzoni et al. report different oscillator frequencies than I have.

Low priority

  • Profile Python's memory usage.

  • SPICE: Add phase via delayed voltage delivery.

  • Plot: Frequency diversity (band) (y-axis) vs. number of oscillators (x-axis) for n algorithms. Hypothesis: As the number of oscillators increases, the frequency diversity increases.

  • Algorithm: Algorithm finds the best combination of parameters iteratively.

  • Implement different signal generation function for Python, more similar to spice (instead of sawtooth).

  • Renaming:

    • algo_monte_carlo -> algo_monte_carlo_greedy
    • algo_mcmc -> algo_monte_carlo_ergodic
    • param_util -> param_funcs
    • param_test_py -> param_py_test
    • param_test_spipy -> param_hybrid_test
    • param -> param_py, param_hybrid
  • Make phase a separate vector of an ensample similar to weights and allow separate phase tuning without replacing the underlying time series

  • solve non-linear transformation (NLT) tasks. for example sinus to square wave.