-
Notifications
You must be signed in to change notification settings - Fork 0
/
sunshine_tools.py
208 lines (178 loc) · 6.45 KB
/
sunshine_tools.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
import numpy as np
import math
from scipy.stats import norm
import bisect
from collections import defaultdict
# group of methods for derivatives
def build_averaged_values_arr(arr, window_size):
n = len(arr)
av = np.zeros(n)
cur_sum = sum(arr[:window_size - 1])
for i in range(window_size - 1, n):
cur_sum += arr[i]
if i - window_size >= 0:
cur_sum -= arr[i - window_size]
av[i] = cur_sum / window_size
return av
def build_averaged_values(df, window_size):
n = df.shape[0]
av = np.zeros(n)
cur_sum = sum(df.loc[:window_size - 2, 'val'])
for i in range(window_size - 1, n):
cur_sum += df.loc[i, 'val']
if i - window_size >= 0:
cur_sum -= df.loc[i - window_size, 'val']
av[i] = cur_sum / window_size
return av
def build_derivatives(df):
n = df.shape[0]
deriv = np.zeros(n)
for i in range(1, n):
if 'year' not in df or df.loc[i, 'year'] == df.loc[i - 1, 'year']:
deriv[i] = df.loc[i, 'val'] - df.loc[i - 1, 'val']
return deriv
def build_relation(df, window_size):
averaged_values = build_averaged_values(df, window_size)[window_size:]
derivatives = np.abs(build_derivatives(df))[1:]
average_derivatives = build_averaged_values_arr(derivatives, window_size)[window_size - 1:]
n = df.shape[0]
result = np.zeros(n)
result[window_size:] = averaged_values / average_derivatives
return result
def make_dates(df):
transformer = lambda dt: datetime.date(int(dt['year']), int(dt['month']), int(dt['day'])).toordinal()
df['date_in_days'] = df.apply(transformer, axis=1)
def plot_relation(df, window_size):
df['relation'] = build_relation(df, window_size)
plt.figure(figsize=(12, 7))
plt.grid(True)
s = r'$Y(t) = \frac{X_T(t)}{\dot{X}_T(t)}$'
s += " window=" + "{:.2f}".format(window_size/365) + " years"
plt.title(s, fontsize=18)
plt.xlabel('Time')
plt.ylabel('Relation')
n = df.shape[0]
years = np.linspace(df.loc[window_size, 'year'], df.loc[n - 1, 'year'], n - window_size)
plt.plot(years, df.loc[window_size:, 'relation'])
def plot_average(df, window_size):
average = build_averaged_values(df, window_size)
plt.figure(figsize=(12, 7))
plt.grid(True)
plt.title('av(X) window=' + "{:.2f}".format(window_size/365) + " years")
plt.xlabel('Time')
plt.ylabel('Average sunshine (hours)')
n = df.shape[0]
years = np.linspace(df.loc[window_size - 1, 'year'], df.loc[n - 1, 'year'], n - window_size + 1)
plt.plot(years, average[window_size-1:])
def plot_deriv(df, window_size):
deriv = abs(build_derivatives(df, window_size))
plt.figure(figsize=(12, 7))
plt.grid(True)
plt.title('|av\'(X)| window=' + "{:.2f}".format(window_size/365) + " years")
plt.xlabel('Time')
plt.ylabel('Abs of derivative')
n = df.shape[0]
years = np.linspace(df.loc[window_size, 'year'], df.loc[n - 1, 'year'], n - window_size)
plt.plot(years, deriv[window_size:])
# group of methods for MRC extraction
def get_binary_map(arr, threshold):
return np.array(arr) >= threshold
def get_entropy_from_counters(counter_1, counter_2, window_size):
entr = 0
for k, v in counter_2.items():
prob = v * 1.0 / (window_size - 1)
if prob > 0:
entr -= math.log(prob, 2) * prob
for k, v in counter_1.items():
prob = v * 1.0 / window_size
if prob > 0:
entr += math.log(prob, 2) * prob
return entr
def get_entropy(arr):
counter = {}
for bin_str in arr:
if bin_str in counter:
counter[bin_str] += 1
else:
counter[bin_str] = 1
entr = 0
for k, v in counter.items():
prob = v / len(arr)
entr -= math.log(prob, 2) * prob
return entr
# get S(p, a) = E(2) - E(1)
def get_entropy_diff(s):
counter = {}
counter1 = [0, 0]
n = len(s)
for i in range(0, n - 1):
if s[i:i+2] not in counter:
counter[s[i:i+2]] = 1
else:
counter[s[i:i+2]] += 1
counter1[int(s[i])] += 1
counter1[int(s[n - 1])] += 1
entr = 0
for k, v in counter.items():
prob = v * 1.0 / (n - 1)
entr -= math.log(prob, 2) * prob
for v in counter1:
prob = v * 1.0 / n
if prob > 0:
entr += math.log(prob, 2) * prob
return entr
def retrive_p_from_data(data, threshold):
data = np.array(data)
mean_ = data.mean()
std_ = data.std()
return 1 - norm.cdf(threshold, loc=mean_, scale=std_)
def get_slope_entropy(p, k=100):
eps = np.linspace(0.01, 1, k)
slope = []
for e in eps:
s = p * (1 - p) * e * (2 * math.log(e, 2) + math.log(p, 2) + math.log(1-p, 2))
s += p * (1 - e + p * e) * math.log(1 - e + p * e, 2)
s += (1 - p) * (1 - p * e) * math.log(1 - p * e, 2)
slope.append(-s)
return eps, slope
def get_exper_entropy(data, threshold, window):
data = np.array(data)
bin_map = get_binary_map(data, threshold).astype(np.int)
n = len(data)
res = []
counter_1 = defaultdict(int)
counter_2 = defaultdict(int)
for i in range(window - 1):
counter_1[str(bin_map[i])] += 1
if i > 0:
counter_2[str(bin_map[i - 1]) + str(bin_map[i])] += 1
for i in range(window - 1, n):
counter_1[str(bin_map[i])] += 1
counter_2[str(bin_map[i - 1]) + str(bin_map[i])] += 1
if i >= window:
counter_1[str(bin_map[i - window])] -= 1
counter_2[str(bin_map[i - window]) + str(bin_map[i - window + 1])] -= 1
e = get_entropy_from_counters(counter_1, counter_2, window)
res.append(e)
return res
def find_lifetime(entropy, entropys_by_p, window):
ans = 0
#print(entropys_by_p[1])
idx = bisect.bisect(entropys_by_p[1], entropy )
if idx == len(entropys_by_p[1]):
print('ERROR: Can`t find epsilon for current entropy {:.5f}'.format(entropy))
idx = len(entropys_by_p[1]) - 1
eps = entropys_by_p[0][idx]
return 1/eps
# returns array of size data.size() - window + 1
def lifetimes_from_data(data, window, threshold):
bin_map = get_binary_map(data, threshold)
p = retrive_p_from_data(data, threshold)
#print('p = ', p)
p = 0.5
entropys_by_p = get_slope_entropy(p)
entropys = get_exper_entropy(data, threshold, window)
#print('Entropys: ', entropys_by_p)
#print('From data: ', entropys)
lifetimes = [find_lifetime(entropy, entropys_by_p, window) for entropy in entropys]
return lifetimes