Skanda319

#we use a psuedo-random number generator
#to pull a random number and determine
#if a server has won a point or not
#theoretically, we could use a more advanced function
#that takes in a distribution and generates random numbers
#based on that
from random import random

#############################################################################################
#define what a big point is and

Skanda319

#cleaned up original code to work with python 3.6
#results match the output from the python 2.7 version
#filter and map functions have been changed between 3.6 and 2.7
import numpy as np
import pandas as pd
import os
import re
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.metrics.pairwise import euclidean_distances
from sklearn.metrics.pairwise import cosine_similarity

Skanda319

#import packages for use later in the HMM code

import pandas as pd
import sklearn.mixture as mix

import numpy as np
import scipy.stats as scs

import datetime as dt

Skanda319

X_train = train[["date", "return", "range", "close"]].set_index("date")
X_test = test[["date", "return", "range", "close"]].set_index("date")

model = mix.GaussianMixture(n_components=3, 
                            covariance_type="full", 
                            n_init=100, 
                            random_state=7).fit(X_train)

# Predict the optimal sequence of internal hidden state
hidden_states = model.predict(X_test)

Skanda319

sns.set(font_scale=1.25)
style_kwds = {'xtick.major.size': 3, 'ytick.major.size': 3,
              'font.family':u'courier prime code', 'legend.frameon': True}
sns.set_style('white', style_kwds)

fig, axs = plt.subplots(model.n_components, sharex=True, sharey=True, figsize=(12,9))
colors = cm.rainbow(np.linspace(0, 1, model.n_components))

for i, (ax, color) in enumerate(zip(axs, colors)):
    # Use fancy indexing to plot data in each state.

Skanda319

sns.set(font_scale=1.5)
states = (pd.DataFrame(hidden_states, columns=['states'], index=X_test.index)
          .join(X_test, how='inner')
          .reset_index(drop=False)
          .rename(columns={'index':'Date'}))
states.head()

#suppressing warnings because of some issues with the font package
#in general, would not rec turning off warnings.
import warnings

Skanda319

import pandas as pd
import io
import requests

url = "https://coinmetrics.io/newdata/"+"btc"+".csv"
s = requests.get(url).content
btc_data = pd.read_csv(io.StringIO(s.decode('utf-8')))

#we add this in to tag a larger data frame if we were
#pulling data for multiple coins

Skanda319

x_cols = ['AdrActCnt', 'BlkCnt', 'BlkSizeByte', 'BlkSizeMeanByte', 'CapMVRVCur',
          'CapMrktCurUSD', 'CapRealUSD', 'DiffMean', 'FeeMeanNtv', 'FeeMeanUSD', 'IssContNtv',
          'IssContPctAnn', 'IssContUSD', 'IssTotNtv', 'IssTotUSD', 'NVTAdj',
          'NVTAdj90','TxCnt', 'TxTfrCnt', 'TxTfrValAdjNtv', 'TxTfrValAdjUSD',
          'TxTfrValMeanNtv', 'TxTfrValMeanUSD', 'TxTfrValMedNtv',
          'TxTfrValMedUSD', 'TxTfrValNtv', 'TxTfrValUSD', 'VtyDayRet180d',
          'VtyDayRet30d', 'VtyDayRet60d']

y_col = ['PriceUSD']

Skanda319

# implements a rolling ridge regression of length window

window = 250 # train window

dt_list = []
pred_list = []
up1sd =[]
down1sd = []
rsq = []