def main():
"""The main function."""
# parse command line arguments
parser = argparse.ArgumentParser()
parser.add_argument('--new', action='store_true')
args = parser.parse_args()
# tells pandas what data types the columns of the .csv file are
dtypes = {
'match_id': int,
'win': bool,
'estimate': int,
'score': float
}
# use new data
if args.new == True:
print('Using new OpenDota api data')
print('Gathering data...' )
# from preprocessing.py
data = get_data(True)
# defaults to old data
else:
print('Using openDota.csv data')
print('Gathering data...' )
# get data from csv file
data = read_csv('data/openDota.csv', dtypes, True)
print('Processing data...')
# process data into required shapes (N, 20) and (N, )
data, labels = process(data)
print("Number of complete matches: " + str(data.shape[0]))
while data.shape[0] < 1000:
print("Not enough matches, trying again")
data = get_data(True)
data, labels = process(data)
best_model = train(data, labels)
print("Beginning testing on new data...")
test(best_model, args.new, dtypes)
# path to save your model
open('best_model.pkl', 'w')
path = os.getcwd() + '/best_model.pkl'
joblib.dump(best_model, path)
def filter_data(news_df):
"""过滤数据"""
df = preprocessing.data_filter(news_df)
now_time = datetime.strftime(datetime.now(), '%Y-%m-%d %H:%M')
# now_time = '2018-04-06 23:59'
df = preprocessing.get_data(df, last_time=now_time, delta=5)
return df
def main():
print(tf.__version__)
print("GPU Available: ", tf.test.is_gpu_available())
results_dict = {'eval_rewards': []}
args = create_argument_parser().parse_args()
segrot, states, markpos = get_data(file=args.expert_file)
actions = get_actions_from_segrot(segrot)
action_dim = actions.shape[1]
state_dim = states.shape[1]
args.action_dim = action_dim
args.state_dim = state_dim
if args.curtail_length:
states = states[0:args.curtail_length + 1]
actions = actions[0:args.curtail_length + 1]
num_states = states.shape[0]
num_train = int(0.9 * num_states)
num_test = num_states - num_train
train_states = states[1:num_train]
train_actions = actions[1:num_train]
test_states = states[-num_test:]
test_actions = actions[-num_test:]
base_dir = os.getcwd() + '/models/IDPAgent/'
run_number = 0
while os.path.exists(base_dir + str(run_number)):
run_number += 1
base_dir = base_dir + str(run_number)
os.makedirs(base_dir)
idp_agent = IDPAgent(**args.__dict__)
for epoch in trange(args.epochs):
train(idp_agent, train_states, train_actions, args.batch_size)
eval_rewards = evaluate_policy(idp_agent, test_states, test_actions,
args.eval_episodes, args.batch_size)
eval_reward = sum(eval_rewards) / args.eval_episodes
eval_variance = float(np.var(eval_rewards))
results_dict['eval_rewards'].append({
'total_steps':
epoch * states.shape[0],
'epoch':
epoch,
'average_eval_reward':
eval_reward,
'eval_reward_variance':
eval_variance
})
with open(args.results_file, 'w') as file:
file.write(json.dumps(results_dict['eval_rewards']))
utils.save_model(idp_agent.actor, base_dir)
def data_filter():
if filter_df.shape[0] == 0:
messagebox.showinfo('Message', '未选择任何新闻数据！')
return
date_f = Entry_Date.get()
day_f = Entry_Day.get()
if date_f == '' or day_f == '':
messagebox.showinfo('Message', '请先填写筛选的日期和天数！')
return
global filter_df0
filter_df0 = preprocessing.get_data(filter_df, last_time=date_f + ' 23:59', delta=int(day_f))
news_pandas.save_news(filter_df0, os.path.join(temp_news_path, 'filter_news_by_time.csv'))
news_num = filter_df0.shape[0]
filter_n.set(news_num)
def test(model, new, dtypes):
'''
Author: Jordan Patterson
Function to test live data on a pretrained model
Parameters
----------
model: Object
Any scikit learn generated model
new: boolean
Whether we trained on new data from OpenDota or not
dtypes: dictionary
Key / value pair specifying type of each column/header
'''
print("Getting test data...")
# don't test on same data we trained on
if new == True:
data = read_csv('data/openDota.csv', dtypes, True)
else:
data = get_data(True)
print("Processing test data...")
data, labels = process(data)
print("Number of complete matches: " + str(data.shape[0]))
# create array of random values
s = np.arange(data.shape[0])
np.random.shuffle(s)
# shuffle data
test_labels = labels[s]
test_data = data[s]
# get test accuracy
acc = model.score(test_data, test_labels) * 100
print("Test accuracy: " + str(acc))
def main():
print(tf.__version__)
print("GPU Available: ", tf.test.is_gpu_available())
args = create_argument_parser().parse_args()
segrot, states, markpos = get_data(file=args.expert_file)
actions = get_actions_from_segrot(segrot)
if args.curtail_length:
states = states[0:args.curtail_length + 1]
actions = actions[0:args.curtail_length + 1]
action_dim = actions.shape[1]
state_dim = states.shape[1]
args.action_dim = action_dim
args.state_dim = state_dim + action_dim
"""
Create environment
"""
env = IDPEnvironment(states[1:], actions[1:], args.max_steps,
args.rand_init)
eval_env = IDPEnvironment(states[1:], actions[1:], args.max_steps,
args.rand_init)
if args.create_testset:
num_states = states.shape[0]
num_train = int(0.9 * num_states)
num_test = num_states - num_train
train_states = states[1:num_train]
train_actions = actions[1:num_train]
test_states = states[-num_test:]
test_actions = actions[-num_test:]
env = IDPEnvironment(train_states, train_actions, args.max_steps,
args.rand_init)
eval_env = IDPEnvironment(test_states, test_actions, args.max_steps,
args.rand_init)
if args.noise == 'ou':
noise = OrnsteinUhlenbeckActionNoise(mu=np.zeros(args.action_dim),
sigma=float(args.noise_scale) *
np.ones(args.action_dim))
elif args.noise == 'normal':
noise = NormalActionNoise(mu=np.zeros(args.action_dim),
sigma=float(args.noise_scale) *
np.ones(args.action_dim))
else:
noise = None
if args.noise == 'param':
param_noise = AdaptiveParamNoiseSpec(
initial_stddev=args.noise_scale,
desired_action_stddev=args.noise_scale)
else:
param_noise = None
base_dir = os.getcwd() + '/models/GACAgent/'
run_number = 0
while os.path.exists(base_dir + str(run_number)):
run_number += 1
base_dir = base_dir + str(run_number)
os.makedirs(base_dir)
gac = GACAgent(**args.__dict__)
state = env.reset()
results_dict = {
'train_rewards': [],
'eval_rewards': [],
'actor_losses': [],
'value_losses': [],
'critic_losses': []
}
episode_steps, episode_rewards = 0, 0 # total steps and rewards for each episode
num_steps = args.num_steps
if num_steps is not None:
nb_epochs = int(num_steps) // (args.epoch_cycles * args.rollout_steps)
else:
nb_epochs = 500
_reset_noise(gac, noise, param_noise)
"""
training loop
"""
average_rewards = 0
count = 0
total_steps = 0
train_steps = 0
for epoch in trange(nb_epochs):
for cycle in range(args.epoch_cycles):
for rollout in range(args.rollout_steps):
"""
Get an action from neural network and run it in the environment
"""
# print('t:', t)
if total_steps < args.start_timesteps:
action = env.sample_action()
else:
action = gac.select_perturbed_action(
tf.convert_to_tensor([state], dtype=tf.float32), noise,
param_noise)
# remove the batch_size dimension if batch_size == 1
action = tf.squeeze(action, [0]).numpy()
# modify action from [-1, 1] to [-180, 180]
next_state, reward, is_terminal = env.step(action)
next_state, reward = np.float32(next_state), np.float32(reward)
gac.store_transition(state, action, reward, next_state,
is_terminal)
episode_rewards += reward
# print('average_rewards:', average_rewards)
# check if game is terminated to decide how to update state, episode_steps,
# episode_rewards
if is_terminal:
state = np.float32(env.reset())
results_dict['train_rewards'].append(
(total_steps, episode_rewards))
episode_steps = 0
episode_rewards = 0
_reset_noise(gac, noise, param_noise)
else:
state = next_state
episode_steps += 1
# evaluate
if total_steps % args.eval_freq == 0:
eval_rewards = evaluate_policy(gac, eval_env,
args.eval_episodes)
eval_reward = sum(eval_rewards) / args.eval_episodes
eval_variance = float(np.var(eval_rewards))
if args.verbose:
results_dict['eval_rewards'].append({
'total_steps':
total_steps,
'train_steps':
train_steps,
'average_eval_reward':
eval_reward,
'eval_reward_variance':
eval_variance,
'eval_rewards_list':
eval_rewards
})
else:
results_dict['eval_rewards'].append({
'total_steps':
total_steps,
'train_steps':
train_steps,
'average_eval_reward':
eval_reward,
'eval_reward_variance':
eval_variance
})
with open(args.results_file, 'w') as file:
file.write(json.dumps(results_dict['eval_rewards']))
total_steps += 1
# train
if gac.replay.size >= args.batch_size:
for _ in range(args.training_steps):
if train_steps % args.param_noise_interval == 0 and param_noise is not None:
episode_transitions = gac.replay.sample_batch(
args.batch_size)
states = episode_transitions.s
unperturbed_actions = gac.get_action(states)
perturbed_actions = episode_transitions.a
ddpg_dist = ddpg_distance_metric(
perturbed_actions.numpy(),
unperturbed_actions.numpy())
param_noise.adapt(ddpg_dist)
gac.train_one_step()
train_steps += 1
with open('results.txt', 'w') as file:
file.write(json.dumps(results_dict))
utils.save_model(gac.actor, base_dir)
from utils.plotting import plot_dataframe_attribute
from utils.preprocessing import get_data, impute_data
if __name__ == "__main__":
data = get_data("../data/titanic3.xls")
imputed_data = impute_data(data)
plot_dataframe_attribute(imputed_data, "age", ["mean", "std"])
from utils.plotting import plot_dataframe_attribute
from utils.preprocessing import get_data, impute_data
if __name__ == "__main__":
data = get_data("../data/titanic3.xls")
imputed_data = impute_data(data)
plot_dataframe_attribute(imputed_data, "age", ["mean", "std"])