In [1]:

    

import numpy as np

from btgym import BTgymEnv
from gym import spaces
import backtrader as bt
from btgym.research import DevStrat_4_11, DevStrat_4_12

%matplotlib inline
import matplotlib.pyplot as plt
plt.style.use('seaborn-white')


    

    




/Users/muzikin/anaconda/envs/tensorforce/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters

In [2]:

    

class SignalProvider():
    
    def __init__(self, env):
        self.env = env
        self.done = True
    
    def get_batch(self, batch_size):
        batch_ext = []
        batch_int = []
        batch_r = []
        while len(batch_r) < batch_size:
            if not self.done:
                o, r, self.done, i = env.step(env.action_space.sample())
            else:
                o = env.reset()
                r = 0
                self.done = False
            batch_ext.append(o['external'])
            batch_int.append(o['internal'])
            batch_r.append(r)
        return batch_ext, batch_int, batch_r
    
    def close(self):
        self.env.close()
        self.done = True


    

In [ ]:

    

# Our Scale-state-before-tanh parameter of interest:
# either a scalar or vector of scaling wieghts, size equal to obs. space num channels:
# here, 6 channels for DevStrat_4_12; 5 for DevStrat_4_11

# Try both to ensure weighted scaling gives much evenly distributed values across channels
# Todo: here weights are linearly decayed. Can do better?

state_ext_scale = np.linspace(3e3, 1e3, num=6)
#state_ext_scale = 3e3

# additional params exist but not used with this example:
state_int_scale = 1.0
reward_scale = 10.0

engine = bt.Cerebro()
engine.addstrategy(
    DevStrat_4_12,
    drawdown_call=10, # max % to loose, in percent of initial cash
    target_call=10,  # max % to win, same
    skip_frame=10,
    gamma=0.99,
    reward_scale=reward_scale,
    state_ext_scale=state_ext_scale,
    #state_int_scale=state_int_scale,
)
env = BTgymEnv(
    filename='../examples/data/DAT_ASCII_EURUSD_M1_2016.csv',
    #filename='../examples/data/DAT_ASCII_EURUSD_M1_201703_1_10.csv',  # ten days
    #filename='../examples/data/test_bent_sine_1min_period1500_300_delta0002.csv',  # increasing sine freq
    #filename='../examples/data/test_sine_1min_period256_delta0002.csv',  # simple sine 
    #state_shape=dict(raw_state=spaces.Box(low=0,high=1,shape=(30,4))),
    engine=engine,
    skip_frame=10,
    port=5050,
    data_port=5049,
    verbose=0
) 
signal_provider = SignalProvider(env)


    

    




/Users/muzikin/anaconda/envs/tensorforce/lib/python3.6/site-packages/matplotlib/__init__.py:1405: UserWarning: 
This call to matplotlib.use() has no effect because the backend has already
been chosen; matplotlib.use() must be called *before* pylab, matplotlib.pyplot,
or matplotlib.backends is imported for the first time.

  warnings.warn(_use_error_msg)

In [ ]:

    

# Make it  256 -:- 1024 (takes a while):
batch_size = 256

batch_ext, _, _ = signal_provider.get_batch(batch_size)
signal_provider.close()


# Distribution of values per channel:
distr_ext = np.asarray(batch_ext)
num_channels = distr_ext.shape[-1]
distr_ext = np.reshape(distr_ext[...,0,:], [-1, num_channels])


for i in range(num_channels):
    plt.figure(num=i, figsize=(10, 5))
    plt.title('External state channel: {}, points: {}'.format(i, distr_ext.shape[0]))
    plt.grid(True)
    fig = plt.hist(distr_ext[:,i], histtype='stepfilled', bins=500)


    

In [5]:

    

signal_provider.close()


    

In [ ]: