Here I'm experimenting with strategies for trading Forex using some ML approaches.
RL Approach contains an archive of a predious attempt, involving tagging winning and losing trades, training a classifier on these trades, then trying to use this to trade successfully.
The more recent attempts have been looking into creating a Signal based model. This model contains four sub-models, each producing entry and exit signals for long and short positions.
Various approaches have been experimented with for training these sub-models, including some traditional supervised learning approaches as well as effective searches over the parameter space; random search, evolutionary search and simulated annealing are all implemented here.
The code first preprocesses tick data from here, .csv files placed in a EURUSD directory. This is then chunked up and used for train/test.
The model preprocesses data it is given by taking diffs between OHLCs, and by using some technical indicators (the parameters for these are included in the model parameter search).