from microprediction import MicroWriter
from microprediction.samplers import exponential_bootstrap
import time
from pprint import pprint
from microprediction.config_private import OXEATE_MAMMAL
# Enters all the z2 streams
WRITE_KEY = OXEATE_MAMMAL
ANIMAL = MicroWriter.animal_from_key(WRITE_KEY)
print(ANIMAL, flush=True)
STOP_LOSS = 250.
if __name__ == "__main__":
mw = MicroWriter(write_key=WRITE_KEY)
LONG = '~' + str(mw.DELAYS[-1])
SHORT = '~' + str(mw.DELAYS[0])
# Give up if things are going badly ... do it nicely
for _ in range(5):
mw.cancel_worst_active(stop_loss=STOP_LOSS, num=1)
time.sleep(1)
for name in mw.get_streams():
if 'z2~' in name and (LONG in name or SHORT in name):
num = mw.num_predictions
z11 = '~'.join(['z1', name.split('~')[1], name.split('~')[3]])
z12 = '~'.join(['z1', name.split('~')[2], name.split('~')[3]])
lagged1 = mw.get_lagged_values(z11)
lagged2 = mw.get_lagged_values(z12)
if len(lagged1) > 10:
# 1. Grab the Github secrets
try:
# For this script to work you need to create four separate GitHub secrets
# called WRITE_KEY_1 WRITE_KEY_2 WRITE_KEY_3 and WRITE_KEY_4
# The idea is that this way you get 900 samples instead of 225
WRITE_KEYS = [os.environ.get('WRITE_KEY_' + str(i + 1)) for i in range(4)]
assert len(WRITE_KEYS) == 4, 'Need four write keys to make the syndicate'
except:
# Or one secret called WRITE_KEY or WRITE_KEYS with them comma separated
WRITE_KEYS_comma_sep = os.environ.get('WRITE_KEYS') or os.environ.get(
'WRITE_KEY')
WRITE_KEYS = WRITE_KEYS_comma_sep.split(',')
print('Copula syndicate is firing up.')
for write_key in WRITE_KEYS:
animal = MicroWriter.animal_from_key(write_key)
print(animal)
# 2. Pick a copula
VINES = [
'center', 'regular', 'direct'
] # See https://sdv.dev/Copulas/tutorials/03_Multivariate_Distributions.html#Vine-Copulas
VINE_TYPE = random.choice(
VINES
) # Perhaps you want to fix this choice. This way we get lots of plots.
# 3. (Optional) Set the URL of your repo so that others can learn from it
REPO = 'https://github.com/microprediction/microactors-plots/blob/master/fit.py' # <--- Change your username
PLOTS_PATH = os.path.dirname(
os.path.realpath(__file__)) + os.path.sep + 'gallery'
from microprediction import MicroWriter
import numpy as np
from pprint import pprint
import matplotlib.pyplot as plt
import random
import time
import warnings
warnings.filterwarnings('ignore')
from copulas.multivariate import GaussianMultivariate
import pandas as pd
# Grab the Github secret
import os
WRITE_KEY = os.environ.get('WRITE_KEY') # <-- You need to add a Github secret
ANIMAL = MicroWriter.animal_from_key(WRITE_KEY) # <-- Your nom de plume
REPO = 'https://github.com/microprediction/microactors/blob/master/fit.py' # <--- Change your username
print('This is ' + ANIMAL + ' firing up')
STOP_LOSS = 25 # <--- Governs when we give up on a stream/horizon
# Get historical data, fit a copula, and submit
def fit_and_sample(lagged_zvalues: [[float]], num: int, copula=None):
""" Example of creating a "sample" of future values
lagged_zvalues: [ [z1,z2,z3] ] distributed N(0,1) margins, roughly
copula : Something from https://pypi.org/project/copulas/
returns: [ [z1, z2, z3] ] representative sample
Swap out this function for whatever you like.
from microprediction import MicroWriter
import time
from pprint import pprint
import numpy as np
# This example illustrates how easy it is to enter z1~ streams
# For a video explanation of z1 streams and see https://www.linkedin.com/feed/update/urn:li:activity:6696569137365430272/
# See https://www.linkedin.com/pulse/short-introduction-z-streams-peter-cotton-phd/ for an explanation of z-streams.
# You need to set a write key.
# Supplying a URL to your repository is optional.
# This will do is make your position on the leaderboards clickable and direct people to your work.
# You can alternatively provide your personal homepage or LinkedIn profile
WRITE_KEY = DODDLE_MAMMAL
ANIMAL = MicroWriter.animal_from_key(DODDLE_MAMMAL)
REPO = 'https://github.com/microprediction/microprediction/blob/master/crawler_alternatives/' + ANIMAL.lower(
).replace(' ', '_') + '.py'
# We simply loop over all z1~ streams
# These streams are approximately normally distributed.
# We enter N(0,1) distributed samples so as to try to take advantage of algorithms
# which accidentally submit distributional predictions far from this.
# This code could be run as a cron job or scheduled task at PythonAnywhere, for example
def jiggle():
return 0.001 * np.random.randn()
if __name__ == "__main__":
