def check_examples():
config = Config("../../../configs/config.json")
size = 120
# pixel_generator = generate_grayscale_pixels(generate_rbg_pixels(config["data_dir"] + "gifs/tenor.gif", window_size=size))
pixel_generator = generate_rbg_pixels(config["data_dir"] +
"gifs/tenor.gif",
window_size=size)
for concurrent_examples in generate_pixel_examples(pixel_generator):
time.sleep(.5)
def text_sequence():
c = Config("../configs/config.json")
data_dir = c["data_dir"] + "Texts/"
input_sequence = sequence_nominal_text(data_dir + "emma.txt")
target_sequence = sequence_nominal_text(data_dir + "emma.txt")
next(target_sequence)
example_sequence = (input_sequence, target_sequence),
return from_sequences(example_sequence)
def check_frames():
config = Config("../../../configs/config.json")
size = 5
original = Image.open(config["data_dir"] + "gifs/tenor.gif")
for no_frame, values in enumerate(
generate_grayscale_pixels(
generate_rbg_pixels(config["data_dir"] + "gifs/tenor.gif",
window_size=size))):
print(f"processing frame {no_frame}...")
write_image(values, original.width // size, original.height // size,
"{:03d}.png".format(no_frame))
def test_nominal_functionality():
# g = env_simple_nominal()
config = Config("../configs/config.json")
g = sequence_nominal_text(config["data_dir"] + "Texts/pride_prejudice.txt")
example_sequence = []
last_shape = next(g)
for this_shape in g:
each_example = last_shape, this_shape
example_sequence.append(each_example)
last_shape = this_shape
print(functionality_nominal(example_sequence))
def test_crypto_linear_functionality():
config = Config("../configs/config.json")
g = sequence_rational_crypto(
config["data_dir"] + "23Jun2017-23Jun2018-1m/EOSETH.csv", 60 * 60 * 24)
examples = []
last_value = next(g)
for this_value in g:
each_example = last_value, this_value - last_value
examples.append(each_example)
last_value = this_value
print(functionality_rational_linear(examples))
pyplot.plot(*zip(*examples))
pyplot.show()
def exchange_rate_sequence(start_stamp: int, end_stamp: int, look_ahead: int,
in_cryptos: Sequence[str], out_crypto: str):
c = Config("../configs/config.json")
data_dir = c["data_dir"] + "binance/"
interval_seconds = 60
inputs = tuple(
sequence_rational_crypto(data_dir + "{:s}ETH.csv".format(_c.upper()),
interval_seconds,
start_val=start_stamp,
end_val=end_stamp) for _c in in_cryptos)
input_sequence = merge_iterators(inputs)
targets = greater_or_equal_than_before(
sequence_rational_crypto(data_dir +
"{:s}ETH.csv".format(out_crypto.upper()),
interval_seconds,
start_val=start_stamp,
end_val=end_stamp), .02, look_ahead)
target_sequence = ((_x, ) for _x in targets)
example_sequences = (input_sequence, target_sequence),
return from_sequences(example_sequences)
# coding=utf-8
from _framework.setup import Setup
from _framework.streams.linear.rational.implementations.exchange_rate_stream import ExchangeRateStream
from _framework.systems.predictors.rational.implementations.rational_average_predictor import RationalAverage
from _framework.systems.predictors.rational.implementations.rational_regression_predictor import RationalLinearRegression
from _framework.systems.predictors.rational.implementations.rational_semiotic_predictor import RationalSemiotic
from tools.load_configs import Config
if __name__ == "__main__":
config = Config("../../configs/config.json")
path_dir = config["data_dir"] + "binance/"
history_length = 1
input_file_paths = path_dir + "EOSETH.csv", path_dir + "SNTETH.csv", path_dir + "QTUMETH.csv", path_dir + "BNTETH.csv"
target_file_paths = path_dir + "BNBETH.csv",
input_dimensions = history_length * len(input_file_paths)
output_dimensions = len(target_file_paths)
exchange_parameters = {
"input_file_paths": input_file_paths,
"target_file_paths": target_file_paths,
"start_time": "2017-08-09T09:00:00+00:00",
"end_time": "2018-07-25T08:30:00+00:00",
"interval_seconds": 60,
"offset_seconds": 60 * 60 * 1,
"history_length": history_length,
}
experiments = (
{
def controlled_grid_interaction(predictor: Predictor,
iterations: int = 500000):
c = Config("../configs/config.json")
data_dir = c["data_dir"] + "grid_worlds/"
grid_world = GridWorldLocal(data_dir + "square.txt", rotational=True)
# grid_world = GridWorldLocal(data_dir + "simple.txt", rotational=False)
# grid_world = GridWorldGlobal(data_dir + "sutton.txt", rotational=False)
controller = SarsaController(grid_world.get_motor_range(),
alpha=.8,
gamma=.1,
epsilon=.1)
# controller = RandomController(grid_world.get_motor_range())
last_sensor = None
last_motor = None
sensor, reward = grid_world.react_to(None)
visualization_steps = iterations // 1000
average_reward = .0
average_error = .0
average_duration = .0
for t in range(iterations):
# get data
this_time = time.time()
concurrent_inputs = (last_sensor, last_motor),
concurrent_outputs = predictor.predict(concurrent_inputs)
concurrent_targets = sensor,
concurrent_examples = (concurrent_inputs[0], concurrent_targets[0]),
predictor.fit(concurrent_examples)
d = time.time() - this_time
# query controller
perception = predictor.get_state(), last_sensor
motor = controller.react_to(perception, reward)
error = sum(
float(_o != _t)
for _o, _t in zip(concurrent_outputs, concurrent_targets)) / len(
concurrent_targets)
average_reward = (average_reward * t + reward) / (t + 1)
average_error = (average_error * t + error) / (t + 1)
average_duration = (average_duration * t + d) / (t + 1)
if (t + 1) % visualization_steps == 0:
Visualize.append("reward", predictor.__class__.__name__,
average_reward)
Visualize.append("error", predictor.__class__.__name__,
average_error)
Visualize.append("duration", predictor.__class__.__name__,
average_duration)
last_sensor = sensor
last_motor = motor
sensor, reward = grid_world.react_to(motor)
if Timer.time_passed(2000):
print("Finished {:05.2f}%...".format(100. * t / iterations))
Visualize.finalize("reward", predictor.__class__.__name__)
Visualize.finalize("error", predictor.__class__.__name__)
Visualize.finalize("duration", predictor.__class__.__name__)
def experiment(iterations: int = 500):
size = 120
out_dim = 1
no_ex = 4
plots = {
"error train": {
RationalSemioticModel.__name__, Regression.__name__,
MovingAverage.__name__
},
"error test": {
RationalSemioticModel.__name__, Regression.__name__,
MovingAverage.__name__
},
"duration": {
RationalSemioticModel.__name__, Regression.__name__,
MovingAverage.__name__
}
}
outputs_train = {
f"output train{_o:02d}/{_e:02d}": {
RationalSemioticModel.__name__, Regression.__name__,
MovingAverage.__name__, "target train"
}
for _o in range(out_dim) for _e in range(no_ex)
}
outputs_test = {
f"output test{_o:02d}/{_e:02d}": {
RationalSemioticModel.__name__, Regression.__name__,
MovingAverage.__name__, "target test"
}
for _o in range(out_dim) for _e in range(no_ex)
}
plots.update(outputs_train)
plots.update(outputs_test)
Visualize.init("gif", plots, x_range=iterations, refresh_rate=40)
config = Config("../configs/config.json")
predictor = RationalSemioticModel(input_dimension=1,
output_dimension=out_dim,
no_examples=no_ex,
alpha=100,
sigma=.5,
drag=100,
trace_length=1)
pixel_generator = generate_grayscale_pixels(
generate_rbg_pixels(config["data_dir"] + "gifs/tenor.gif",
window_size=size))
train_streams = generate_pixel_examples(pixel_generator)
test_streams = generate_pixel_examples(pixel_generator)
setup(predictor, train_streams, test_streams, 1, iterations=iterations)
for _each_output in outputs_train:
Visualize.finalize(_each_output, "target train")
for _each_output in outputs_test:
Visualize.finalize(_each_output, "target test")
print("Generating regression model...")
predictor = Regression(input_dimension=1,
output_dimension=out_dim,
drag=100,
no_examples=no_ex)
pixel_generator = generate_grayscale_pixels(
generate_rbg_pixels(config["data_dir"] + "gifs/tenor.gif",
window_size=size))
train_streams = generate_pixel_examples(pixel_generator)
test_streams = generate_pixel_examples(pixel_generator)
setup(predictor, train_streams, test_streams, 1, iterations=iterations)
for _each_output in outputs_train:
Visualize.finalize(_each_output, "target train")
for _each_output in outputs_test:
Visualize.finalize(_each_output, "target test")
print("Generating average model...")
predictor = MovingAverage(output_dimension=out_dim,
drag=100,
no_examples=no_ex)
pixel_generator = generate_grayscale_pixels(
generate_rbg_pixels(config["data_dir"] + "gifs/tenor.gif",
window_size=size))
train_streams = generate_pixel_examples(pixel_generator)
test_streams = generate_pixel_examples(pixel_generator)
setup(predictor, train_streams, test_streams, 1, iterations=iterations)
for _each_output in outputs_train:
Visualize.finalize(_each_output, "target train")
for _each_output in outputs_test:
Visualize.finalize(_each_output, "target test")
print("done!")
Visualize.show()