reward_transformer = load(os.path.join("../../models/reward_transformer.pkl"))
testDf = mod_df.tail(120000).head(100000)
composite_env = CompositeEnv(testDf,
openerPriceDiffGenerator,
buyer_feat_gen,
sellerPriceDiffGenerator,
startDeposit=300,
lot_size=0.1,
lot_coef=100000,
spread=18,
spread_coef=0.00001,
stopType="const",
takeType="const",
stopPos=2,
takePos=2,
maxLoss=20000,
maxTake=20000,
stoplossPuncts=200,
takeprofitPuncts=100,
riskPoints=110,
riskLevels=5,
parallelOpener=False,
renderDir="../../data/pictures",
renderName="test_plot",
env_name="test_env",
turn_off_spread=False)
buyer_env = composite_env.state_dict["buyer"]
openerAgent = DQNAgent("opener", composite_env.observation_space["opener"],
composite_env.action_space["opener"].n)
buyerAgent = DQNAgent("buyer", composite_env.observation_space["buyer"],
################################
price_diff_generator = FeatGen_ScaledWindow(feat_list, n_points=256, flat_stack=False)
opener_price_diff_generator = price_diff_generator
buyer_price_diff_generator = price_diff_generator
seller_price_diff_generator = price_diff_generator
#reward_transformer = RewardTransformer()
#reward_transformer.fit(df, spread_coef=0.00001, lot_coef=100000, lot_size=0.1, window_width=40)
#save(reward_transformer, path=os.path.join("../../models/reward_transformer.pkl"))
reward_transformer = load( os.path.join("../../models/reward_transformer.pkl") )
test_df = mod_df.tail(380000).head(40000)
#back_test_df = mod_df.head(50000).tail(3192).head(1192)
test_env = CompositeEnv(test_df, opener_price_diff_generator, buyer_price_diff_generator, seller_price_diff_generator,
start_deposit=300, lot_size=0.1, lot_coef=100000, spread=18, spread_coef=0.00001,
stop_type="const", take_type="const", stop_pos=2, take_pos=2, max_loss=20000, max_take=20000,
stoploss_puncts=2000, takeprofit_puncts=2000, risk_points=110, risk_levels=5, parallel_opener=False,
render_dir="../../data/pictures", render_name="test_plot", env_name="test_env", turn_off_spread=False)
scores = []
for i in range( 10 ):
agent = load( os.path.join("../../models/", "composite_agent_{}.pkl".format(i)))
agent_scores = []
for j in range( 10 ):
score, deals_statistics = agent.use_agent(test_env, render_deals=False)
median_deal = np.median( deals_statistics )
agent_scores.append( median_deal )
agent_mean_score = np.mean( agent_scores )
scores.append( agent_mean_score )
print( "Scores: {}".format( scores ) )
reward_transformer = load(os.path.join("../../models/reward_transformer.pkl"))
trainDf = mod_df.tail(120000).head(100000)
trainEnv = CompositeEnv(trainDf,
openerPriceDiffGenerator,
buyerPriceDiffGenerator,
sellerPriceDiffGenerator,
startDeposit=300,
lot_size=0.1,
lot_coef=100000,
spread=18,
spread_coef=0.00001,
stopType="const",
takeType="const",
stopPos=2,
takePos=1,
maxLoss=20000,
maxTake=20000,
stoplossPuncts=200,
takeprofitPuncts=100,
riskPoints=110,
riskLevels=5,
parallelOpener=False,
renderFlag=True,
renderDir="../../data/pictures",
renderName="train_plot",
env_name="train_env",
turn_off_spread=True)
backTestDf = mod_df.tail(140000).head(20000)
#backTestDf = modDf.head(50000).tail(3192).head(1192)