Slight modification of Facebook Salina Reinforcement Learning - A2C GPU example for financial series.

The gym FOREX data are provided by gym_anytrading library GitHub Page

With respect to the traditional CartPole-V0 gym, the stock_func is designed to provide in input a FOREX trading gym:

def stock_func(max_episode_steps,seed=123, window_size=10, size_sample=100):
    df =  FOREX_EURUSD_1H_ASK.copy()
    start_index = window_size
    if size_sample < 0:              ### put size_sample=-1 to consider the whole dataset.
        end_index = len(df)
    else:
        end_index = size_sample
    env = TimeLimit(gym.make('forex-v0', df=df, window_size=window_size, frame_bound=(start_index, end_index)), max_episode_steps=max_episode_steps)
    env.seed(seed)
    return env

A double tensor with diff close and relative gains are given in the gen_state function. The latter transforms the 'env/obs' tensors collected from AutoResetGymAgent into suitable tensors for Policy/Critic agent neural networks.

def _gen_state(observation):
    index = torch.tensor([0])
    diff_close = torch.transpose(torch.index_select(observation[0], 1, index), 1, 0)
    index2 = torch.tensor([1])
    buy_sell = torch.transpose(torch.index_select(observation[0], 1, index2), 1, 0)
    observation = torch.squeeze(torch.stack([diff_close, buy_sell], dim=1))
    return  observation

I take no responsibility for the use of the code. It is a simple test of SALINA's potential for financial problems.

The software license remains the one indicated in the source code and respectively linked to the official Facebook SALINA repository GitHub Page

Francesco Bardozzo, PhD fbardozzo@unisa.it - NeuroneLab - University of Salerno - Italy

Roberto Tagliaferri, Full Prof. robtag@unisa.it - NueroneLab - University of Salerno - Italy