-
Notifications
You must be signed in to change notification settings - Fork 6
/
fit_large_MNL_LCM.py
242 lines (199 loc) · 9.12 KB
/
fit_large_MNL_LCM.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
#!/usr/bin/env python
# coding: utf-8
from urbansim.models.regression import RegressionModel
import os
import numpy as np
import pandas as pd
import orca
from urbansim.models import util
import yaml
from urbansim_templates.models import LargeMultinomialLogitStep
from urbansim_templates import modelmanager as mm
# from guppy import hpy; h=hpy()
# import pymrmr
# suppress sklearn warnings
def warn(*args, **kwargs):
pass
os.chdir("/home/da/semcog_urbansim")
def columns_in_vars(vars):
hh_columns, b_columns = [], []
for varname in vars:
if ':' in varname:
vs = varname.split(':')
hh_columns.append(vs[0])
b_columns.append(vs[1])
else:
hh_columns.append(varname)
b_columns.append(varname)
return hh_columns, b_columns
def load_hlcm_df(hh_var, b_var):
import models
orca.add_injectable('year', 2020)
buildings = orca.get_table("buildings")
households = orca.get_table("households")
orca.run(["build_networks_2050"])
orca.run(["neighborhood_vars"])
# load both hh
hh = households.to_frame(hh_var)
b = buildings.to_frame(b_var)
return hh, b
hh_sample_size = 10000
estimation_sample_size = 50
job_sample_size = 1000
job_estimation_sample_size = 80
# LARGE_AREA_ID = 147
# number_of_vars_to_use = 40
choice_column = "building_id"
hh_filter_columns = ["building_id", "large_area_id", "mcd_model_quota", "year_built", "residential_units"]
job_filter_columns = ["building_id", "slid", "home_based_status"]
b_filter_columns = ["large_area_id", "mcd_model_quota", "residential_units", "non_residential_sqft", "vacant_job_spaces"]
building_hh_capacity_col = 'residential_units'
building_job_capacity_col = 'vacant_job_spaces'
# reload variables?
def load_hh_and_b(LARGE_AREA_ID=5, RELOAD=False):
if RELOAD:
thetas = pd.read_csv("out_theta_%s_%s.txt" % (LARGE_AREA_ID, estimation_sample_size), index_col=0)
# config
# load variables
# orca.run(["mcd_hu_sampling"])
# TODO: get vars from vars list from last forecast
used_vars = thetas.index
hh_columns, b_columns = columns_in_vars(used_vars)
hh_var = hh_columns + hh_filter_columns
b_var = b_columns + b_filter_columns
hh_region, b_region = load_hlcm_df(hh_var, b_var)
hh_region.to_csv('hh.csv')
b_region.to_csv('b.csv')
else:
hh_region = pd.read_csv('hh.csv', index_col=0)
b_region = pd.read_csv('b.csv', index_col=0)
return hh_region, b_region
def run_large_MNL(hh_region, b_region, LARGE_AREA_ID, number_of_vars_to_use=40):
thetas = pd.read_csv("./configs/hlcm_2050/thetas/out_theta_%s_%s.txt" % (LARGE_AREA_ID, estimation_sample_size), index_col=0)
hh = hh_region[hh_region.large_area_id == LARGE_AREA_ID]
hh = hh[hh.building_id > 1]
hh = hh[hh.residential_units > 0]
hh = hh[hh.year_built > 2005]
# exclude hh in pseudo buildings
hh = hh[hh.building_id < 90000000]
hh = hh.fillna(0) # found 12 missing values in ln_income
hh = hh[[col for col in hh.columns if col not in hh_filter_columns+["household_id"]]+['building_id']]
b = b_region[b_region.large_area_id == LARGE_AREA_ID]
b = b[b.residential_units > 0]
b = b[[col for col in b.columns if col not in b_filter_columns or col == building_hh_capacity_col]]
# remove extra columns
hh_cols_to_std = [col for col in hh.columns if col not in ['building_id']]
# standardize hh
hh[hh_cols_to_std] = (hh[hh_cols_to_std]-hh[hh_cols_to_std].mean())/hh[hh_cols_to_std].std()
# building columsn to standardize
b_cols_to_std = [col for col in b.columns if col not in [building_hh_capacity_col]]
# standardize buildings
b[b_cols_to_std] = (b[b_cols_to_std]-b[b_cols_to_std].mean())/b[b_cols_to_std].std()
# fillin na with 0
b = b.fillna(0)
# filter out columns with 0 std
b_cols_with_0_std = b.columns[b.std()==0]
# adding hh and b to orca
orca.add_table('hh_hlcm', hh)
orca.add_table('b_hlcm', b)
m = LargeMultinomialLogitStep()
m.choosers = ['hh_hlcm']
m.chooser_sample_size = min(hh_sample_size, hh.shape[0])
# m.chooser_filters = chooser_filter
# Define the geographic alternatives agent is selecting amongst
m.alternatives = ['b_hlcm']
m.choice_column = choice_column
m.alt_sample_size = estimation_sample_size
# m.alt_filters = alts_filter
# use top k variables
# filter variables
# some variables has 0 std, need to remove them for the MNL to run
v = thetas.theta.abs().sort_values(ascending=False).index
v_wo_0_std = [col for col in v if all(
[vv.strip() not in b_cols_with_0_std for vv in col.split(':')])]
selected_variables = v_wo_0_std[:number_of_vars_to_use]
m.model_expression = util.str_model_expression(selected_variables, add_constant=False)
m.constrained_choices = True
m.alt_capacity = building_hh_capacity_col
m.fit()
m.name = 'hlcm_%s' % (LARGE_AREA_ID)
mm.register(m)
print('done')
def run_elcm_large_MNL(job_region, b_region, SLID, number_of_vars_to_use=40):
print("Running elcm largeMNL on slid %s..." % SLID)
thetas = pd.read_csv("./configs/elcm_2050/thetas/out_theta_job_%s_%s.txt" % (SLID, job_estimation_sample_size), index_col=0)
job = job_region[job_region.slid == SLID]
job = job[job.building_id > 1]
job = job[job.home_based_status == 0]
job = job.fillna(0) # found 12 missing values in ln_income
job = job[[col for col in job.columns if col not in job_filter_columns+["job_id"]]+['building_id']]
b = b_region[b_region.large_area_id == SLID % 1000]
b = b[b.non_residential_sqft > 0]
b = b[[col for col in b.columns if col not in b_filter_columns or col == building_job_capacity_col]]
# remove extra columns
job_cols_to_std = [col for col in job.columns if col not in ['building_id', building_job_capacity_col]]
# standardize job
job[job_cols_to_std] = (job[job_cols_to_std]-job[job_cols_to_std].mean())/job[job_cols_to_std].std()
b_cols_to_std = [col for col in b.columns if col not in [building_job_capacity_col]]
# standardize buildings
b[b_cols_to_std] = (b[b_cols_to_std]-b[b_cols_to_std].mean())/b[b_cols_to_std].std()
# fillin na with 0
b = b.fillna(0)
# filter out columns with 0 std
b_cols_with_0_std = b.columns[b.std()==0]
# adding job and b to orca
orca.add_table('job_elcm', job)
orca.add_table('b_elcm', b)
m = LargeMultinomialLogitStep()
m.choosers = ['job_elcm']
m.chooser_sample_size = min(job_sample_size, job.shape[0])
# m.chooser_filters = "(slid==%s) & (building_id>1) & (home_based_status==0)" % (SLID)
# m.chooser_sample_size = min(job_sample_size, jobs.query(m.chooser_filters).shape[0])
# Define the geographic alternatives agent is selecting amongst
m.alternatives = ['b_elcm']
m.choice_column = choice_column
m.alt_sample_size = job_estimation_sample_size
# m.alt_filters = "(large_area_id==%s) & (non_residential_sqft>0)" % (SLID % 100)
# use top k variables
# filter variables
# some variables has 0 std, need to remove them for the MNL to run
v = thetas.theta.abs().sort_values(ascending=False).index
v_wo_0_std = [col for col in v if all(
[vv.strip() not in b_cols_with_0_std for vv in col.split(':')])]
# remove vacant_job_spaces from selected variables
v_wo_0_std = [col for col in v_wo_0_std if col not in ['vacant_job_spaces']]
selected_variables = v_wo_0_std[:number_of_vars_to_use]
# remove variables which with correlation close to 1
i = number_of_vars_to_use
while ((b[selected_variables].corr()>0.99).sum()>1).sum() > 0 and i<len(v_wo_0_std):
corr = b[selected_variables].corr()
corr_ge_1_sum = (corr>0.99).sum()
# drop the first variable
var_to_drop = corr_ge_1_sum[corr_ge_1_sum>1].index[0]
print("Removed %s from expression due to close to 1 correlation" % var_to_drop)
selected_variables.remove(var_to_drop)
print("Added %s as replacement" % v_wo_0_std[i])
selected_variables.append(v_wo_0_std[i])
i += 1
m.model_expression = util.str_model_expression(selected_variables, add_constant=False)
m.constrained_choices = True
m.alt_capacity = building_job_capacity_col
m.fit()
fp = m.model.results['fit_parameters']
rho = [m.model.results['log_likelihood']['rho_bar_squared']]
while fp[(fp['P-Values']>0.1) | (fp['P-Values'].isna())].shape[0]>0 and fp.shape[0] > 2:
# when there are variables with p > .1
# removing them and refit
new_vars_inds = [i for i in range(
fp.shape[0]) if i not in fp[(fp['P-Values']>0.1) | (fp['P-Values'].isna())].index]
new_vars = [m.model.results['x_names'][i] for i in new_vars_inds]
m.model_expression = util.str_model_expression(new_vars, add_constant=False)
m.fit()
fp = m.model.results['fit_parameters']
rho.append(m.model.results['log_likelihood']['rho_bar_squared'])
m.name = 'elcm_%s' % (SLID)
mm.register(m)
print('done')
if __name__ == '__main__':
hh_region, b_region = load_hh_and_b(93, False)
run_large_MNL(hh_region, b_region, 93, 60)