-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
177 lines (147 loc) · 6.66 KB
/
model.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
import networkx as nx
from scipy import sparse
import numpy as np
from math import cos, sin, pi
from glob import glob
import numpy as np
import scipy
from numpy.linalg import inv
from scipy.stats import multivariate_normal
from tqdm import tqdm
import scipy.sparse as sp
import pickle
import random
def lhsu(xmin,xmax,nsample):
nvar = len(xmin)
ran = np.random.rand(nsample,nvar)
s = np.zeros((nsample,nvar))
for j in range(nvar):
idx = list(range(nsample))
random.shuffle(idx)
idx = np.array(idx)
P = (idx-ran[:,j])/nsample;
s[:,j] = xmin[j] + np.dot(P,xmax[j]-xmin[j])
s = s.T
return s
def checkbound(x,xmin,xmax):
n = x.shape[0]
for i in range(n):
x[i,:][x[i,:] < xmin[i]] = xmin[i]
x[i,:][x[i,:] > xmax[i]] = xmax[i]
return x
def fill_in_taus(A, P, H, L, taus):
tau_p2p, tau_p2h, tau_p2l, tau_h2p, tau_h2h, tau_h2l, tau_l2p, tau_l2h = taus
R = A.tocoo()
R.setdiag(0)
R.sum_duplicates()
R.eliminate_zeros()
for idx, i, j in zip(range(R.nnz), R.row, R.col):
if i < P and j < P:
R.data[idx] *= tau_p2p
elif i < P and j < P+H:
R.data[idx] *= tau_h2p
elif i < P and j < P+H+L:
R.data[idx] *= tau_l2p
elif i < P+H and j < P:
R.data[idx] *= tau_p2h
elif i < P+H and j < P+H:
R.data[idx] *= tau_h2h
elif i < P+H and j < P+H+L:
R.data[idx] *= tau_l2h
elif i < P+H+L and j < P:
R.data[idx] *= tau_p2l
elif i < P+H+L and j < P+H:
R.data[idx] *= tau_h2l
else:
R.data[idx] *= 0
return R.tocsr()
def simulate_subroutine(H_now, H_t_now, P, H, L, taus, loads, states, alpha, g_threshold):
N = P + H + L
## TODO
## 1) Aggregate patient, HCW and Location loads in parallel using the various taus
###### REMEMBER TO USE THE RIGHT TAUS, we have three for each node type
## 2) Sum up over vertices
## 3) Elementwise product with H with calculated loads
## 4) Apply g over each hyperedge and sum up over hyperedges (Use tanh and g_threshold)
#print(f"new_loads shape is {new_loads.shape}")
#print(f"states shape is {states.shape}")
#print(f"P is {P}")
#print(f"H shape is {H.get_shape()}")
out_pp = H_now[:, :P] * taus['P']['P']
out_ph = H_now[:, P:P+H] * taus['P']['H']
out_pl = H_now[:, P+H:] * taus['P']['L']
patient_out = np.squeeze(np.asarray(sparse.hstack([out_pp, out_ph, out_pl]).sum(axis=1)))
out_hp = H_now[:, :P] * taus['H']['P']
out_hh = H_now[:, P:P+H] * taus['H']['H']
out_hl = H_now[:, P+H:] * taus['H']['L']
hcw_out = np.squeeze(np.asarray(sparse.hstack([out_hp, out_hh, out_hl]).sum(axis=1)))
out_lp = H_now[:, :P] * taus['L']['P']
out_lh = H_now[:, P:P+H] * taus['L']['L']
out_ll = H_now[:, P+H:] * taus['L']['L']
location_out = np.squeeze(np.asarray(sparse.hstack([out_lp, out_lh, out_ll]).sum(axis=1)))
patient_outs = np.squeeze(np.asarray(H_t_now[:P, :].multiply(patient_out).sum(axis=1)))
hcw_outs = np.squeeze(np.asarray(H_t_now[P:P+H, :].multiply(hcw_out).sum(axis=1)))
location_outs = np.squeeze(np.asarray(H_t_now[P+H:, :].multiply(location_out).sum(axis=1)))
outs = np.concatenate((patient_outs, hcw_outs, location_outs))
new_loads = loads * (1-outs)
h_l = H_now.multiply(loads).tocsr()
#print(f"h_l shape is {h_l.shape}")
#print(f"N is {N}")
# big_matrix = sparse.csc_matrix((H.get_shape()[0],N), dtype='float32')
# Patients Receiving
l_pp = h_l[:, :P] * taus['P']['P']
l_ph = h_l[:, P:P+H] * taus['P']['H']
l_pl = h_l[:, P+H:] * taus['P']['L']
patient_sum = np.squeeze(np.asarray(np.tanh(sparse.hstack([l_pp, l_ph, l_pl]).sum(axis=1))*g_threshold))
#patient_sum = np.squeeze(np.asarray(sparse.hstack([l_pp, l_ph, l_pl]).sum(axis=1)))
#print(patient_sum)
# HCW Receiving
l_hp = h_l[:, :P] * taus['H']['P']
l_hh = h_l[:, P:P+H] * taus['H']['H']
l_hl = h_l[:, P+H:] * taus['H']['L']
hcw_sum = np.squeeze(np.asarray(np.tanh(sparse.hstack([l_hp, l_hh, l_hl]).sum(axis=1))*g_threshold))
#hcw_sum = np.squeeze(np.asarray(sparse.hstack([l_hp, l_hh, l_hl]).sum(axis=1)))
# Location Receiving
l_lp = h_l[:, :P] * taus['L']['P']
l_lh = h_l[:, P:P+H] * taus['L']['H']
l_ll = h_l[:, P+H:] * taus['L']['L']
location_sum = np.squeeze(np.asarray(np.tanh(sparse.hstack([l_lp, l_lh, l_ll]).sum(axis=1))*g_threshold))
#location_sum = np.squeeze(np.asarray(sparse.hstack([l_lp, l_lh, l_ll]).sum(axis=1)))
# TODO: For further optimization, don't need to use variables
# Sum up loads for each node type into the big matrix
#print(f"h_l shape is {h_l.shape}")
#print(f"patient_sum shape is {patient_sum.shape}\n{patient_sum}")
#print(f"big_matrix shape is {big_matrix.shape}")
#print(f"h_l.T*patient_sum shape is {h_l.T.multiply(patient_sum).T.shape}")
### TODO ###
### SERIOUSLY RETHINK THIS IMPLEMENTATION, converting a bit toooo much ####
### USE LINE PROFILER ###
patient_loads = np.squeeze(np.asarray(H_t_now[:P, :].multiply(patient_sum).sum(axis=1)))
hcw_loads = np.squeeze(np.asarray(H_t_now[P:P+H, :].multiply(hcw_sum).sum(axis=1)))
location_loads = np.squeeze(np.asarray(H_t_now[P+H:, :].multiply(location_sum).sum(axis=1)))
## WILL NEED TO CHANGE THIS TO ACCOUNT FOR LOCATIONS IN MODEL WITH MORE LOCATIONS
#location_loads = np.array([0.0]*(L+1))
#print(f"patient_loads shape is {patient_loads.shape}\n{patient_loads}")
#print(f"hcw_loads shape is {hcw_loads.shape}\n{hcw_loads}")
#print(f"location_loads shape is {location_loads.shape}\n{location_loads}")
#print(f"H_t.shape is {H_t.shape}")
new_loads += np.concatenate((patient_loads, hcw_loads, location_loads))
return new_loads
def simulate_hyper_weekly(Hs, H_ts, P, H, L, taus, loads, states, parameters, t):
# Hs: List of Hypergraphs. Format is (H,V) where H is the hypergraph and V is the list of vertices/nodes
# H_ts = Hs.T
alpha, beta, delta, g_threshold = parameters[3], parameters[4], parameters[5], parameters[-1]
cases = []
for time in range(7):
rands = np.random.rand(P)
s2i = np.where((rands < beta*loads[:P]) & (states == 0))[0]
i2s = np.where((rands < delta) & (states == 1))[0]
if s2i.size != 0:
states[s2i] = 1
if i2s.size != 0:
states[i2s] = 0
new_loads = simulate_subroutine(Hs[time], H_ts[time], P, H, L, taus, loads, states, alpha, g_threshold)
loads = new_loads.copy()
loads[:P] += alpha*states
cases.append(len(np.where(states == 1)[0]))
return states, loads, cases