Skip to content

Latest commit

 

History

History
48 lines (24 loc) · 1.47 KB

TODO.md

File metadata and controls

48 lines (24 loc) · 1.47 KB

THE TODO LIST:

  • (DONE) Factor standardization use scikit learn

  • (DONE) Change Y values to shifted X values for sequence learning

  • (DONE) Does MSE need to be divided by batch size batch? No. Done in TF.

  • (DONE) Implement Clairvioant and Naive models

  • (DONE) Implement batch sequences that only require final step being an active stock.

  • (DONE) Implement predicting next n-timestep average of inputs in batch_generator

  • (DONE) Incorporate merge-model-with-simdata.pl into euclid2

  • (DONE) Simulate clairvoyant progression from 0,3,6,12, ... months to be how perf improves

  • (DONE) In predict.py, make predictions even when there is no target data available

  • (DONE) Create file cache (pickle) for batch_generator

  • (DONE) Add auxilary input features -- ones that are not predicted/targets (e.g., momentum)

  • (DONE) Re-working scaling/unscaling implementation so it is more intuitive

  • (DONE) Layer normalization in RNN

  • (DONE) In predict.py, output predictions timesteps less than t. I.e., t-1, t-2, 0

  • (DONE) Implement variable length sequences

  • (DONE) RNN cost function upweight last k time steps instead of just last time step

  • Configurable validation/holdout set methodology (holdout time window or companies)

  • Implement a genetic aglorithm for hyper-parameter space search

  • max-norm regularization for RNN and MLP (use tf.clip_by_norm)

  • Trainable ReLu units in MLP

  • rename config's nn_type to model_type

  • Documentation. Starting with README.md

  • Make caching faster