def test_days_in_range(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
sim = Simulator(TSLA, portfolio, path, training)
# span a weekend
gen = sim._days_in_range(datetime(2020, 7, 10), datetime(2020, 7, 14))
self.assertEqual(2, len(list(gen)))
def test_simulate_range(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
sim = Simulator(TSLA, portfolio, path, training)
gen = sim._days_in_range('2020-6-1', '2020-7-31')
self.assertEqual(pd.Timestamp('2020-6-2'), next(gen)['date'])
self.assertEqual(36, len(list(gen)))
def test_close_short_call(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
call = Option('call', TSLA, 420, datetime(2020, 7, 17))
portfolio = Portfolio(cash=133535, securities={call: -1})
sim = Simulator(TSLA, portfolio, path, training)
sim._buy(datetime(2020, 7, 14))
self.assertEqual(0, portfolio.securities[call])
self.assertEqual(0, portfolio.cash)
def test_close_no_op(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
sim = Simulator(TSLA, portfolio, path, training)
# no contracts to close, no-op
sim._buy(None)
self.assertEqual(100, portfolio.securities[TSLA])
def test_calculate_fitness(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
call = Option('call', TSLA, 420, datetime(2020, 9, 11))
portfolio = Portfolio(cash=10, securities={TSLA: 100, call: -1})
sim = Simulator(TSLA, portfolio, path, training)
actual = sim._calculate_fitness(0.2990607757568724,
datetime(2020, 9, 8)) # 420.28
# starting cash + value of shares - call obligation - buy-and-hold value
expected = 10 + (420.28 * 100) - (1.44 * 100) - (420.28 * 100)
self.assertAlmostEqual(expected, actual, places=2)
def test_expire_without_close_on_expiration(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
call = Option('call', TSLA, 2800, datetime(2020, 7, 24))
portfolio = Portfolio(cash=0, securities={TSLA: 100, call: -1})
portfolio.collateral = {TSLA: 100}
sim = Simulator(TSLA, portfolio, path, training)
net = BuyAndHoldNet()
sim.simulate(net, datetime(2020, 7, 23), datetime(2020, 8, 3))
self.assertEqual(0, portfolio.securities.get(call, 0))
def test_hold_only(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
sim = Simulator(TSLA, portfolio, path, training)
net = BuyAndHoldNet()
fitness = sim.simulate(net, datetime(2020, 7, 20),
datetime(2020, 8, 22))
# close on 8/21 was 2049.98
# buy-and-hold is the fitness base-line
self.assertAlmostEqual(0, fitness, places=3)
def test_open_short_call(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
sim = Simulator(TSLA, portfolio, path, training)
sim._sell(
datetime(2020, 7, 20),
0.293070701634208, # normalized close for 7/20
0.4617712204180135, # should target the 2000 strike
0.3087888191319731 # should target 8/21 expiration
)
call = Option('call', TSLA, 2000, datetime(2020, 8, 21))
self.assertEqual(-1, portfolio.securities[call])
self.assertEqual(100, portfolio.securities[TSLA])
self.assertEqual(100, portfolio.collateral[TSLA])
self.assertEqual(10875, portfolio.cash)
def test_plot_trades(self):
net = RandomNet(5)
reporter = TradeReporter()
path = Path('tests') / 'test_data' / 'normalized' / TSLA.symbol
portfolio = Portfolio(cash=0, securities={TSLA: 100})
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
sim = Simulator(TSLA, portfolio, path, training, reporter)
dr = DateRangeFactory(training).random_date_range(90)
vis.plot_trades(net, sim, dr, training, path, reporter, False)
def eval_genomes(genomes, config):
# all genomes should be compared against using the same date range
t_start, t_end = training_daterange_factory.random_date_range(90)
c_start, c_end = cv_daterange_factory.random_date_range(90)
return_dict = Manager().dict()
for genome_id, genome in genomes:
net = neat.nn.FeedForwardNetwork.create(genome, config)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
t_sim = Simulator(TSLA, portfolio, path, training)
c_sim = Simulator(TSLA, portfolio, path, validation)
train_p = Process(target=worker, args=('training', t_sim, net, t_start, t_end, return_dict))
validation_p = Process(target=worker, args=('validation', c_sim, net, c_start, c_end, return_dict))
train_p.start()
validation_p.start()
train_p.join()
validation_p.join()
genome.fitness = return_dict['training']
genome.cv_fitness = return_dict['validation']
def test_sell_and_expire(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
sim = Simulator(TSLA, portfolio, path, training)
net = SellOnceNet()
# close on 8/21 was 2049.98
# direction,expiration,strike,price,iv,delta,theta,vega
# call,200821,2050.0,100.0,0.49691698172618826,0.4607403125936822,0.3105944110992971,0.5219995298370175
net.theta = 0.3087888191319731 # target 8/21 expiration
net.delta = 0.4607403125936822 # target 2050 strike
fitness = sim.simulate(net, datetime(2020, 7, 19),
datetime(2020, 8, 22))
# premium + held shares value - buy-and-hold value
expected = (100 * 100) + (2049.98 * 100) - (2049.98 * 100)
self.assertAlmostEqual(expected, fitness, places=2)
self.assertEqual(100, portfolio.securities[TSLA])
self.assertEqual(0, portfolio.collateral[TSLA])
self.assertEqual(10000, portfolio.cash)
def test_sell_and_get_assigned(self):
path = Path('tests/test_data/normalized/TSLA')
training = pd.read_csv(path / 'training.csv',
parse_dates=['date'],
date_parser=from_small_date)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
sim = Simulator(TSLA, portfolio, path, training)
net = SellOnceNet()
# close on 8/21 was 2049.98
# direction,expiration,strike,price,iv,delta,theta,vega
# call,200821,2000.0,108.75,0.4965503408681624,0.46177122041801355,0.30752325473559794,0.5241433005010039
net.theta = 0.3087888191319731 # target 8/21 expiration
net.delta = 0.4617712204180135 # target 2000 strike
fitness = sim.simulate(net, datetime(2020, 7, 19),
datetime(2020, 8, 22))
# premium + assigned cash - buy-and-hold value
expected = (108.75 * 100) + (2000 * 100) - (2049.98 * 100)
self.assertAlmostEqual(expected, fitness, places=2)
self.assertEqual(0, portfolio.securities[TSLA])
self.assertEqual(0, portfolio.collateral[TSLA])
self.assertEqual(10875 + 200000, portfolio.cash)
def run(config_file):
config = neat.Config(
neat.DefaultGenome,
neat.DefaultReproduction,
neat.DefaultSpeciesSet,
neat.DefaultStagnation,
config_file
)
while True:
checkpoint = sorted(glob('neat-checkpoint-*'), reverse=True)
if checkpoint:
pop = neat.Checkpointer.restore_checkpoint(checkpoint[0])
else:
pop = neat.Population(config)
stats = neat.StatisticsReporter()
# add a stdout reporter to show progress in terminal
pop.add_reporter(neat.StdOutReporter(True))
pop.add_reporter(stats)
pop.add_reporter(neat.Checkpointer(50))
winner = pop.run(eval_genomes, 50)
# display the winning genome
print(f"\nBest genome:\n{winner}")
win_net = neat.nn.FeedForwardNetwork.create(winner, config)
view = False
vis.plot_stats(stats, ylog=False, view=view)
vis.plot_species(stats, view=view)
portfolio = Portfolio(cash=0, securities={TSLA: 100})
simulator = Simulator(TSLA, portfolio, path, training)
daterange = training_daterange_factory.random_date_range(90)
reporter = TradeReporter()
vis.plot_trades(win_net, simulator, daterange, training, path, reporter, view=view)
node_names = {0: 'Buy', 1: 'Sell', 2: 'Hold', 3: 'Delta', 4: 'Theta'}
vis.draw_net(config, winner, view=view, node_names=node_names)