-
Notifications
You must be signed in to change notification settings - Fork 1
/
MarkovChainMonteCarlo.py
255 lines (221 loc) · 8.28 KB
/
MarkovChainMonteCarlo.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
"""
Author: Christian Lozoya, 2017
"""
import os
import random as random
import re
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from sklearn.externals import joblib
random.seed(0)
np.random.seed(0)
ID = 'STRUCTURE_NUMBER_008'
LATITUDE = 'LAT_016'
LONGITUDE = 'LONG_017'
ENCODING = 'latin1'
NA_VALUES = ['N']
def read_folder(directory, ext):
"""
Collect all file paths with given extension (ext) inside directory
"""
paths = []
for file in os.listdir(directory):
if file.split('.')[-1].lower() == ext.lower():
paths.append(file)
return sorted(paths)
def process_data(item, directory, ext, clean=False):
"""
item is the item of interest (str)
dir is the directory where the files of data are found (str)
paths is all the paths of data in the directory (list of str)
clean determines whether or not incomplete data is removed
returns a list of pandas dataFrames and a list of numerics to name the files
"""
data = []
yrs = []
paths = read_folder(directory, ext)
for i, file in enumerate(paths):
data.append(pd.read_csv(directory + '/' + file, usecols=[ID] + [item], na_values=NA_VALUES,
dtype={**{ID: str}, **{item: str}}, encoding=ENCODING))
yrs.append(re.findall(r"[0-9]{4}(?=.txt)", file)[0])
data[i].set_index(ID, inplace=True)
data[i] = data[i][~data[i].index.duplicated()]
data = concat_data(dataFrame=data, columns=yrs, index=item)
if clean:
data = clean_data(dataFrame=data, columns=yrs)
return data, yrs
def concat_data(dataFrame, columns, index=None):
"""
Concatenate list of pandas Series into a DataFrame
dataFrame is a pandas DataFrame and columns are the columns in data
index is used for concatenating only specific columns
"""
if index:
cnc = pd.concat([d[index] for d in dataFrame], axis=1)
else:
cnc = pd.concat([d for d in dataFrame], axis=1)
cnc.columns = columns
return cnc
def count_matrix(data, matrix, columns=None, hst=False, offset=None):
"""
Count data and store the count in matrix (pre-labeled DataFrame objects)
columns is a list of column labels
"""
data = data.astype('int8')
if not hst:
matrix.fillna(0, inplace=True)
length = len(data.columns)
for i, row in enumerate(data.iterrows()):
for j, column in enumerate(data.loc[row[0], :]):
if hst:
matrix[data.iloc[i, j]].append(int(columns[j]) - (offset - 1))
elif columns: # If columns are specified: row=the value in (i,j), column=column j
matrix.loc[data.iloc[i, j], columns[j]] += 1
else: # If nothing is specified: transition matrix is assumed
if (j < length - 1):
try:
if (int(data.iloc[i, j + 1]) <= int(data.iloc[i, j])):
matrix.loc[data.iloc[i, j], data.iloc[i, j + 1]] += 1
except Exception as e:
print(e)
return matrix
def normalize_rows(dataFrame):
"""
Normalize dataFrame by row
"""
try:
nrm = dataFrame.div(dataFrame.sum(axis=1), axis=0)
nrm.fillna(0, inplace=True)
return nrm
except Exception as e:
print(e)
def frq(focus, matrix):
"""
Count state transitions in data and store the count in matrix (pre-labeled DataFrame objects)
"""
matrix = matrix
if focus == "year":
frq = matrix.T
frq = frq[frq.columns[::-1]]
return normalize_rows(frq).as_matrix()
elif focus == "state":
return normalize_rows(matrix.T).T.as_matrix()
else:
print("Select focus")
def hst(data, matrix, columns): # TODO bad method + needs generalizing (only works for freq vs state)
print(data)
print(matrix)
hst = count_matrix(data=data, matrix=matrix, columns=columns, hst=True)
hst = pd.DataFrame([hst[m] for m in hst], index=[m for m in hst])
print(hst)
hst.fillna(0, inplace=True)
hst.sort_index(ascending=False, inplace=True)
return hst.as_matrix().astype(np.float64)
def clean_data(dataFrame, columns):
"""
Clears incomplete data resulting in regular matrix
data is a pandas DataFrame and columns are the columns in data
"""
cln = dataFrame
for column in columns:
cln = (cln[np.isfinite(cln[column].astype(float))])
cln.columns = columns
return cln
def get_transition_matrix(data, vS, offset=None):
return normalize_rows(count_matrix(data, pd.DataFrame(index=vS, columns=vS), offset=offset))
def sample(simulation, vS, pV, iteration):
"""
simulation: list
vS: list of valid state names
pV: probability vector
iteration: current iteration, int
returns a state vector
"""
r = random.uniform(0, 1)
if r > 2:
state = np.argmax(pV)
simulation[iteration].append(state)
return simulation, np.transpose((pd.Series([1 if s == state else 0 for s in vS], index=vS)))
else:
sum = 0
randomNumber = random.uniform(0, 1)
# Assign state if randomNumber is within its range
for state, prob in zip(vS, pV):
sum += prob
if (sum >= randomNumber):
if state > simulation[-1][-1]:
simulation[iteration].append(simulation[-1][-1])
return simulation, np.transpose((pd.Series([1 if s == simulation[-1][-1] else 0 for s in vS], index=vS)))
else:
simulation[iteration].append(state)
return simulation, np.transpose((pd.Series([1 if s == state else 0 for s in vS], index=vS)))
def markov_chain_monte_carlo(time, matrix, vS, iS, iterations):
simulation = []
for i in range(iterations):
simulation.append([])
cS = iS
for t in range(time):
if t != 0:
# Multiply current state and transition matrix, resulting in a probability vector
pV = cS.dot(np.linalg.matrix_power(matrix, t))
simulation, cS = sample(simulation, vS, pV, i)
else:
for state, prob in zip(vS, cS):
if prob == 1:
simulation[i].append(state)
simulation[i] = pd.Series(simulation[i])
return simulation
def model_monte_carlo(time, model, vS, iS, iterations):
simulation = []
age = np.array([time])
kind = np.array([0 if i != 1 else 1 for i in range(9)])
type = np.array([0 if i != 19 else 1 for i in range(22)])
for i in range(iterations):
simulation.append([])
cS = iS
for t in range(time):
#print(np.argmax(cS))
if t != 0:
cS = np.hstack([age, np.argmax(cS), kind, type])
pV = model.predict_proba([cS])[0]
simulation, cS = sample(simulation, vS, pV, i)
else:
for state, prob in zip(vS, cS):
if prob == 1:
simulation[i].append(state)
simulation[i] = pd.Series(simulation[i])
return simulation
plt.ylim(0, 9)
dataDir = 'training'
item = 'DECK_COND_058'
iterations = 10000
clean = True
nStates = 10
offset = 1992
data, yrs = process_data(item=item, directory=dataDir, ext='txt', clean=clean)
o = data.T.astype('int8')
o.index = o.index.astype(int) - offset
#plt.plot(o.mode(axis=1), linestyle=':')
plt.plot(o.mean(axis=1), linestyle='--')
vS = tuple(range(10))
time = len(data.columns)
#Q = get_transition_matrix(data, vS, offset=offset)
model = joblib.load("model.joblib.dat")
plt.xlabel('Time')
plt.ylabel('Condition')
plt.title('{} Gradient Boost Simulations'.format(iterations))
for initialState in range(9, nStates):
iS = np.array([0 if i != initialState else 1 for i in range(nStates)])
#simulationA = markov_chain_monte_carlo(100, Q, vS, iS, iterations=iterations)
#c = pd.concat(simulationA, axis=1)
#plt.plot(c.mode(axis=1), linestyle=':')
#plt.plot(c, linewidth=0.5)
#plt.plot(c.mean(axis=1), linewidth=2.0)
simulationB = model_monte_carlo(100, model, vS, iS, iterations=iterations)
c = pd.concat(simulationB, axis=1)
#plt.plot(c.mode(axis=1))
plt.plot(c, linestyle=':', linewidth=0.5)
plt.plot(c.mean(axis=1), linewidth=2.0, label='Simulations Mean')
plt.legend(loc='best')
plt.show()