def main(train_stock, val_stock, window_size, batch_size, ep_count, model_name,
pretrained, debug):
""" Trains the stock trading bot using Deep Q-Learning.
Please see https://arxiv.org/abs/1312.5602 for more details.
Args: optional arguments [python train.py --help]
"""
switch_k_backend_device()
agent = Agent(window_size, pretrained=pretrained, model_name=model_name)
train_data = get_stock_data(train_stock)
val_data = get_stock_data(val_stock)
initial_offset = val_data[1] - val_data[0]
for episode in range(1, ep_count + 1):
train_result = train_model(agent,
episode,
train_data,
ep_count=ep_count,
batch_size=batch_size,
window_size=window_size)
val_result, _ = evaluate_model(agent, val_data, window_size, debug)
show_train_result(train_result, val_result, initial_offset)
def main(eval_stock, window_size, model_name, debug):
""" Evaluates the stock trading bot.
Please see https://arxiv.org/abs/1312.5602 for more details.
Args: optional arguments [python evaluate.py --help]
"""
switch_k_backend_device()
data = get_stock_data(eval_stock)
initial_offset = data[1] - data[0]
if model_name is not None:
'''Single Model Evaluation'''
agent = Agent(window_size, pretrained=True, model_name=model_name)
profit, _ = evaluate_model(agent, data, window_size, debug)
show_eval_result(model_name, profit, initial_offset)
del agent
else:
'''Multiple Model Evaluation'''
for model in os.listdir('models'):
if not os.path.isdir('models/{}'.format(model)):
agent = Agent(window_size, pretrained=True, model_name=model)
profit = evaluate_model(agent, data, window_size, debug)
show_eval_result(model, profit, initial_offset)
del agent
min_sup = int(args[args.index("-s") + 1])
except:
min_sup = int(args[args.index("--min-sup") + 1])
if "-c" in args or "--min-conf" in args:
try:
min_conf = float(args[args.index("-c") + 1])
except:
min_conf = float(args[args.index("--min-conf") + 1])
# Check if result directory exists, if it doesn't make one
if not os.path.isdir("results"):
os.mkdir("results")
# Download stock data
if not get_stock_data(ticker):
print(
f"Data for ${ticker} has previously been fetched, skipping fetching..."
)
# Parse stock data into event sequence
print("[!] Converting csv data into a sequence...")
time_series = get_time_series(ticker)
word_length = int(word_length * len(time_series)) if word_length else int(
word_length_multiplier * len(time_series))
print("[!] Generating event sequence...")
event_sequence = convert_to_event_sequence(time_series, word_length,
alphabet_size)
def test_get_comment_counts(self): data = get_stock_data() counts = get_comment_counts(data) self.assertTrue(isinstance(counts, dict)) self.assertTrue(isinstance(counts["AAPL"], int)) self.assertTrue(isinstance(counts["GOOG"], int))
'''Faster computation on CPU (only if using tensorflow-gpu)'''
if TENSORFLOW_BACKEND:
print('Switching to TensorFlow for CPU...')
os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
if len(sys.argv) == 8:
train_stock_name, val_stock_name, window_size, batch_size, episode_count, model_name, pretrained = sys.argv[1], \
sys.argv[2], int(sys.argv[3]), int(sys.argv[4]), int(sys.argv[5]), sys.argv[6], bool(int(sys.argv[7]))
else:
print(
'Usage: python train.py [train stock] [val stock] [window] [batch size] [episodes] [model] [pretrained (0/1)]'
)
exit(0)
agent = Agent(window_size, pretrained=pretrained, model_name=model_name)
train_data = get_stock_data(train_stock_name)
test_data = get_stock_data(val_stock_name)
initial_offset = test_data[1] - test_data[0]
for episode in range(1, episode_count + 1):
train_result = train_model(agent,
episode,
train_data,
episode_count=episode_count,
batch_size=batch_size,
window_size=window_size)
val_result = evaluate_model(agent, test_data, window_size=window_size)
show_train_result(train_result, val_result, initial_offset)
print('Done Training!')
model_name = None
if len(sys.argv) == 3:
stock_name, window_size = sys.argv[1], int(sys.argv[2])
elif len(sys.argv) == 4:
stock_name, window_size, model_name = sys.argv[1], int(
sys.argv[2]), sys.argv[3]
else:
print('Usage: python evaluate.py [stock] [window] [model (optional)]')
print(
'NOTE - All models in "models/" dir will be evaluated if no pretrained model is provided'
)
exit(0)
data = get_stock_data(stock_name)
initial_offset = data[1] - data[0]
if model_name is not None:
'''Single Model Evaluation'''
agent = Agent(window_size, pretrained=True, model_name=model_name)
profit = evaluate_model(agent, data, window_size=window_size)
show_eval_result(model_name, profit, initial_offset)
del agent
else:
'''Multiple Model Evaluation'''
for model in os.listdir('models'):
if not os.path.isdir('models/{}'.format(model)):
agent = Agent(window_size, pretrained=True, model_name=model)
profit = evaluate_model(agent, data, window_size=window_size)
show_eval_result(model, profit, initial_offset)