def fbprophet_univariate_best_r() -> float:
"""
:return: Provides best r for a randomly chosen data stream
Takes about 12 hrs to run
"""
mr = MicroReader()
names = mr.get_stream_names()
okay = False
while not okay:
name = random.choice(names)
n_obs = len(mr.get_lagged_values(name=name))
okay = n_obs > PROPHET_META['n_warm'] + 50 and '~' not in name
url = 'https://www.microprediction.org/stream_dashboard.html?stream=' + name.replace(
'.json', '')
print('We will find the best fbprophet hyper-parameters for ' + url)
print('There are ' + str(n_obs) + ' observations in the series.')
print(
"Prophet will be fit for most of them, after a burn_in, and for many different hyper-params. Don't hold your breathe."
)
best_r, best_value, info = optimal_r_for_stream(
f=fbprophet_univariate_r2,
name=name,
k=10,
optimizer=dlib_default_cube,
n_burn=PROPHET_META['n_warm'] + 20,
n_trials=50,
n_dim=2)
pprint(info)
params = prophet_params(r=best_r, dim=2)
pprint(params)
def stream_data(name: str, n_obs: int):
""" values and times for a univariate stream """
mr = MicroReader()
lagged_values, lagged_times = mr.get_lagged_values_and_times(
name=name, count=n_obs)
y, t = list(reversed(lagged_values)), list(reversed(lagged_times))
return y, t
def random_regular(min_len=500):
""" Randomly selected univariate series
:return: y, t, url
"""
mr = MicroReader()
names = mr.get_stream_names()
okay = False
while not okay:
name = random.choice(names)
n_obs = len(mr.get_lagged_values(name=name,count=10000))
okay = n_obs > min_len and '~' not in name
url = 'https://www.microprediction.org/stream_dashboard.html?stream=' +name.replace('.json','')
return name, url
def next_opinionated_forecast(n_train, k, n_recent, multiple, name=None):
while True:
try:
from microprediction import MicroReader
n_obs = 0
while n_obs < 1000:
mr = MicroReader()
if name is None:
names = mr.get_stream_names()
random_name = random.choice(names)
else:
random_name = name
lag_values, lag_times = mr.get_lagged_values_and_times(
name=random_name, count=2000)
y = list(reversed(lag_values))
t = list(reversed(lag_times))
n_obs = len(y)
except ImportError:
from timemachines.skatertools.data import hospital
y = hospital()
t = [15 * 60 * i for i in range(len(y))]
name = 'hospital'
for i in [100 * j for j in range(10)]:
print('Looking at ' + random_name + ' ' + str(i) + '/1000')
if len(y) > i + 2 * k + n_train:
y_fit = y[i:i + n_train]
t_fit = t[i:i + n_train]
y_hats, _, forecast, m = prophet_iskater_factory(y=y_fit,
k=k,
t=t_fit)
if is_opinonated(y=y_fit,
forecast=forecast,
k=k,
n_recent=n_recent,
multiple=multiple):
y_3avg = np.mean(
y[i + n_train - 3:i +
n_train]) # avg of last three points
y_true_mean = np.mean(
y[i + n_train + k - 1:i + n_train + k +
1]) # avg of 3 future points
error = (y_hats[-1] - y_true_mean) / abs(0.01 + y_3avg)
avg_error = (y_3avg - y_true_mean) / abs(0.01 + y_3avg)
return forecast, m, random_name, error, avg_error, y[
i + n_train:i + n_train + k]
print(random_name + ' is okay.')
return None, None, None
def random_stream_name(min_len=500, exclude_str=None, include_str=None):
""" Randomly selected univariate series
:return: y, t, url
"""
mr = MicroReader()
names = mr.get_stream_names()
okay = False
while not okay:
name = random.choice(names)
n_obs = len(mr.get_lagged_values(name=name, count=10000))
okay = True
if exclude_str is not None and exclude_str in name:
okay = False
if include_str is not None and include_str not in name:
okay = False
if n_obs < min_len:
okay = False
url = 'https://www.microprediction.org/stream_dashboard.html?stream=' + name.replace(
'.json', '')
return name, url
def optimal_r_for_stream(
f,
name: str,
k: int,
evaluator=None,
optimizer=None,
n_trials=None,
n_dim=None,
n_burn: int = None,
test_objective_first=True
) -> (float, float, dict): # best_r, best_val, info
""" Find the best hyper-parameters for a univariate skater using live from www.microprediction.org
:param f:
:param name: Choose from https://www.microprediction.org/browse_streams.html but add '.json' to the end
:param k:
:param evaluator:
:param optimizer:
:param n_trials:
:param n_dim:
:param n_burn:
:param test_objective_first:
:return: best_r, best_value, info
"""
mr = MicroReader()
lagged_values, lagged_times = mr.get_lagged_values_and_times(name=name)
t = list(reversed(lagged_times))
y = list(reversed(lagged_values))
return optimal_r(f=f,
y=y,
k=k,
a=None,
t=t,
e=None,
evaluator=evaluator,
optimizer=optimizer,
n_trials=n_trials,
n_dim=n_dim,
n_burn=n_burn,
test_objective_first=test_objective_first)
from microprediction import MicroReader
import random
# Live tests
mr = MicroReader(base_url='https://devapi.microprediction.org')
TEST_STREAM = random.choice(list(mr.get_streams().items()))[0]
DIE = 'die.json'
print(TEST_STREAM)
def test_getters():
""" Not a great test :) """
assert mr.get_current_value(DIE)
assert len(mr.get_lagged_values(DIE)) > 10
for delay in mr.DELAYS:
p1 = mr.get_discrete_pdf_lagged(name=DIE, delay=delay)
assert p1 is not None
if p1.get('x'):
assert abs(p1['y'][0] - 0.1666
) < 0.1, "Oh man this die market is so inefficient!"
def test_z_getters():
zs = mr.get_lagged_zvalues(name='z2~copula_x~copula_y~70.json', count=3)
ps = mr.get_lagged_copulas(name='z2~copula_x~copula_y~70.json', count=3)
assert len(zs) == 3
assert len(zs[0]) == 2
assert len(ps) == 3
assert len(ps[0]) == 2
assert ps[0][0] <= 1
assert ps[0][0] >= 0
def plot_helicopter_lags():
""" Plot a subset of the SciML helicopter challenge data .. psi only """
mr = MicroReader()
xs = mr.get_lagged_values('helicopter_psi.json')
plt.plot(xs)
# A collection of mostly standalone functions illustrating copula functionality at dev.microprediction.org
# See also https://github.com/microprediction/PDCI/blob/master/helicopula.ipynb
# See article at https://www.linkedin.com/pulse/helicopulas-peter-cotton-phd/
import pandas as pd
import matplotlib.pyplot as plt
from microprediction import MicroReader, MicroWriter, new_key
from copulas.multivariate import GaussianMultivariate
import numpy as np
from pprint import pprint
mr = MicroReader()
def get_wind_z2():
""" Retrieve bivariate wind data lagged values """
HELISTREAM = 'z2~seattle_wind_direction~seattle_wind_speed~70.json'
lagged_values = mr.get_lagged_values(name=HELISTREAM)
return lagged_values
def plot_helicopter_data():
""" Plot SciML helicopter challenge data """
pd.read_csv(
'https://raw.githubusercontent.com/SciML/HelicopterSciML.jl/master/data/Lab-Helicopter_Experimental-data.csv'
).plot()
def plot_helicopter_lags():
""" Plot a subset of the SciML helicopter challenge data .. psi only """
def test_internet():
mr = MicroReader()
streams = mr.get_sponsors()
assert len(streams) > 10
def test_imports():
mr = MicroReader()
values = mr.get_lagged_values(name='cop.json')
assert len(values)>500
from microprediction.univariate.expnormdist import ExpNormDist, DEFAULT_EXPNORM_LOWER
import os
import numpy as np
from copy import deepcopy
import time
from microprediction import MicroReader
mr = MicroReader()
STREAMS = mr.get_stream_names()
DEFAULT_EXPNORM_PARAMS = {'g1': 0.5, 'g2': 5.0, 'logK': -2., 'loc': 0.0, 'logScale': 0.0}
DEFAULT_EXPNORM_LOWER = {'g1': 0.001, 'g2': 0.001, 'logK': -5, 'loc': -0.15, 'logScale': -4}
DEFAULT_EXPNORM_UPPER = {'g1': 1.0, 'g2': 15.0, 'logK': 1, 'loc': 0.15, 'logScale': 4.0}
OFFLINE_EXPNORM_HYPER = {'lower_bounds': deepcopy(DEFAULT_EXPNORM_LOWER),
'upper_bounds': deepcopy(DEFAULT_EXPNORM_UPPER),
'space': None, 'algo': None, 'max_evals': 3}
class ExpNormAccumulator(ExpNormDist):
def __init__(self, **kwargs):
super().__init__(**kwargs)
def anchors(self, lagged_values, lagged_times):
def post_getter(state,value):
return state['anchor']
return self.replay(lagged_values=lagged_values, lagged_times=lagged_times, post_getter=post_getter)
def manual_loss(self, lagged_values, lagged_times, params, state, burn_in=10):
""" Loss function for a series of values, calculated manually as check """
def test_delays_1():
mc = MicroReader()
assert len(mc.DELAYS) == 3
from microprediction import MicroReader
from pprint import pprint
import numpy as np
import matplotlib.pyplot as plt
def to_density(cdf):
""" """
# CDF is a fast, noisy O(1) approximation so this isn't the greatest
dys = np.diff([0] + cdf['y'])
dxs = np.diff([cdf['x'][0] - 1.0] + cdf['x'])
dsty = [dy / dx for dx, dy in zip(dxs, dys)]
return [d / sum(dsty) for d in dsty]
if __name__ == "__main__":
mr = MicroReader()
HOUR = mr.DELAYS[3]
cdf = mr.get_cdf(name='altitude.json', delay=HOUR)
# plt.plot(cdf['x'], cdf['y'])
plt.plot(cdf['x'], to_density(cdf))
plt.show()
print(
'https://www.microprediction.org/stream_dashboard.html?stream=altitude.json'
)