-
Notifications
You must be signed in to change notification settings - Fork 0
/
hada_omlt.py
564 lines (476 loc) · 27.9 KB
/
hada_omlt.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
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
import argparse
import time
import pandas as pd
import pickle
import warnings
import os
import csv
import pyomo.environ as pyo
from pyomo.contrib.appsi.base import TerminationCondition
from omlt import OmltBlock
from omlt import scaling
from omlt.gbt.gbt_formulation import GBTBigMFormulation, add_formulation_to_block
from omlt.gbt.model import GradientBoostedTreeModel
from sklearn.ensemble import GradientBoostingRegressor
from skl2onnx import convert_sklearn
from skl2onnx.common.data_types import FloatTensorType
from prettytable import PrettyTable
from ml import train_quality_GBT, train_targets_GBT
warnings.filterwarnings('ignore')
####### METADATA ########
HW = ["pc", "vm", "g100"]
TARGET = ["time", "quality", "memory", "price"]
ALG_PARAMS = {
"convolution" : 4,
"saxpy" : 3,
#"blackscholes" : 15,
"correlation" : 7,
"fwt" : 2
}
####### UTILITY FUNCTIONS #######
def logger(x, export_log):
if export_log:
with open("hada.log", "a") as file_log:
file_log.write(f"{x}\n")
print(x)
def extract_var_bounds(data_folder, algorithm, price):
'''
Compute upper and lower bounds of each variable
PARAMETERS
---------
algorithm [string]: a transprecision computing algorithm {saxpy, convolution, correlation, fwt}
data_folder [string]: path to the folder containing the necessary datasets
price [dict]: price of each hw platform
RETURN
------
var_bounds [pd.DataFrame]: a frame with lower/upper bound for each variable
'''
bounds_min = {}
bounds_max = {}
for hw in HW:
dataset = pd.read_csv(f"{data_folder}/{algorithm}_{hw}.csv")
bounds_min[hw] = dataset.min()
bounds_max[hw] = dataset.max()
var_bounds = pd.DataFrame({
"min" : pd.DataFrame(bounds_min).transpose().min(),
"max" : pd.DataFrame(bounds_max).transpose().max()
})
dataset = pd.read_csv(f"{data_folder}/{algorithm}_quality.csv")
var_bounds.loc['quality', 'min'] = dataset['quality'].min()
var_bounds.loc['quality', 'max'] = dataset['quality'].max()
for i in range(ALG_PARAMS[algorithm]):
var_bounds.loc["var_" + str(i), "max"] = 53
var_bounds = var_bounds.append(pd.DataFrame(
{"min": [min(price.values())], "max": [max(price.values())]},
index = ["price"]))
print(var_bounds)
return var_bounds
def extract_robust_coeff(robust_factor, data_folder, algorithm, mlmodel_files, TARGET):
'''
Compute robustness coefficients for each predictive model, according to the specified robustness factor
PARAMETERS
---------
robust_factor [float]: confidence margin to introduce into the user constraints
data_folder [string]: path to the folder containing the necessary datasets
algorithm [string]: a transprecision computing algorithm {saxpy, convolution, correlation, fwt}
mlmodel_files [dict]: path to the predictive model corresponding to each pair (hw, target)
TARGET [list]: targets specified by the user
RETURN
------
robust_coeff [dict]: robustness coefficient for each predictive model
'''
if robust_factor or robust_factor == 0:
robust_coeff = {}
for target in TARGET:
# The target quality requires a single robustness coefficient
if target == "quality":
dataset = pd.read_csv(f"{data_folder}/{algorithm}_quality.csv")
model = pickle.load(open(mlmodel_files[(f"{target}")], 'rb'))
dataset[f'{target}_pred'] = model.predict(dataset[[col for col in dataset.columns if 'var' in col]])
dataset[f'{target}_error'] = (dataset[f'{target}'] - dataset[f'{target}_pred']).abs()
robust_coeff[(f"{target}")] = dataset[f'{target}_error'].std() * dataset[f'{target}_error'].quantile(robust_factor)
else:
for hw in HW:
# The target price is not estimated: it does not require any robustness coefficient
if target == 'price':
robust_coeff[(f"{hw}", "price")] = 0
else:
dataset = pd.read_csv(f"{data_folder}/{algorithm}_{hw}.csv")
model = pickle.load(open(mlmodel_files[(f"{hw}", f"{target}")], 'rb'))
dataset[f'{target}_pred'] = model.predict(dataset[[col for col in dataset.columns if 'var' in col]])
dataset[f'{target}_error'] = (dataset[f'{target}'] - dataset[f'{target}_pred']).abs()
robust_coeff[(f"{hw}", f"{target}")] = dataset[f'{target}_error'].std() * dataset[f'{target}_error'].quantile(robust_factor)
return robust_coeff
else:
return None
def compute_mean_std(data_folder, algorithm, target, hw):
dataset = pd.read_csv(f"{data_folder}/{algorithm}_{target if target == 'quality' else hw}.csv")
mean_data = dataset.mean(axis=0)
std_data = dataset.std(axis=0)
y_mean = mean_data[target]
y_std = std_data[target]
X_mean = mean_data[[f"var_{i}" for i in range(ALG_PARAMS[algorithm])]]
X_std = std_data[[f"var_{i}" for i in range(ALG_PARAMS[algorithm])]]
return X_mean, X_std, y_mean, y_std
def get_scaling(algorithm, target, hw):
'''Computes input bounds and the scaler object
to feed the omlt model'''
X_mean, X_std, y_mean, y_std = compute_mean_std("datasets", algorithm, target, hw)
scaler = scaling.OffsetScaling(offset_inputs=X_mean.tolist(),
factor_inputs=X_std.tolist(),
offset_outputs=[y_mean],
factor_outputs=[y_std])
input_bounds = {i:
((var_bounds.loc[f"var_{i}", "min"] - X_mean.tolist()[i]) / X_std.tolist()[i],
(var_bounds.loc[f"var_{i}", "max"] - X_mean.tolist()[i]) / X_std.tolist()[i]) for i in range(ALG_PARAMS[algorithm])}
return scaler, input_bounds
########### HADA #############
def HADA(algorithm, objective, user_constraint, price, var_bounds, mlmodel_files, export_log=False, robust_coeff=None):
'''
Implement HADA:
1. Declare variables and basic constraints
2. Embed predictive models
3. Declare user-defined constraints and objective
4. Solve the model and output an optimal matching (hw-platform, alg-configuration)
PARAMETERS
---------
algorithm [string] : a transprecision computing algorithm {saxpy, convolution, correlation, fwt}
objective [dict]: type {min, max} and target
user_constraint [dict]: type {'leq', 'geq', 'eq'} and right-hand side of each user-defined constraint
price [dict]: price of each hw platform
var_bounds [pd.DataFrame]: lower and upper bound for each target
mlmodel_files [dict]: path to the predictive model corresponding to each pair (hw, target)
export_log [bool]: when True, the log is written to hada.log
robust_coeff [dict] : robustness coefficient for each pair (hw, target), for each performance target
constrained by the user
RETURN
------
sol [dict]: optimal solution found
model [pyomo.Model]: final optimization model
'''
logger("\n==============> Building", export_log)
####### MODEL #######
start_time = time.time()
# ALL_VARS = [f"b[{hw}]" for hw in HW] + \
# [f"y[{hw}_{target}]" for hw in HW for target in TARGET if target != "quality"] + [f"y[quality]"] + \
# [f"var[{i}]" for i in range(ALG_PARAMS[algorithm])]
DECLARED_VARS = []
model = pyo.ConcreteModel()
####### VARIABLES #######
# Binary variable indexed by hw index set
model.index_b = pyo.Set(initialize=HW)
model.b = pyo.Var(model.index_b, domain=pyo.Binary)
DECLARED_VARS += [f"b[{hw}]" for hw in HW]
# Continuous variables representing the target values
target_index_set = []
for target in set(list(user_constraint.keys()) + [objective['target']]):
if target == "quality":
target_index_set.append('quality')
DECLARED_VARS.append("y[quality]")
else:
for hw in HW:
target_index_set.append(f"{hw}_{target}")
DECLARED_VARS.append(f"y[{hw}_{target}]")
model.index_y = pyo.Set(initialize=target_index_set)
# for indexed variables, bounds are defined through a function
# that has access to the model and indexes
def bound_target_var(model, i):
if i != "quality":
_, i = i.split("_")
return (var_bounds.loc[f"{i}", "min"], var_bounds.loc[f"{i}", "max"])
model.y = pyo.Var(model.index_y, bounds=bound_target_var, domain=pyo.Reals)
# An integer variable for each integer parameter of the algorithm,
# representing the value assigned to this parameter
def bound_algorithm_var(model, i):
return (var_bounds.loc[f"var_{i}", "min"], var_bounds.loc[f"var_{i}", "max"])
model.index_var = pyo.Set(initialize=range(ALG_PARAMS[algorithm]))
model.var = pyo.Var(model.index_var ,domain=pyo.Integers, bounds=bound_algorithm_var)
model.aux_var = pyo.Var(model.index_var ,domain=pyo.Reals, bounds=bound_algorithm_var)
DECLARED_VARS += [f"var[{i}]" for i in range(ALG_PARAMS[algorithm])]
####### CONSTRAINTS ######
# HW Selection Constraint, enabling the selection of a single hw platform
# model.index_hw_selection = pyo.Set(initialize=["hw_selection"])
model.hw_selection = pyo.Constraint(expr=sum(model.b[hw] for hw in model.b) == 1)
# Integrality Constraints, enabling the conversion of the auxiliary variables from continuous to integer
model.eq_constraint = pyo.Constraint(model.index_var, rule=lambda model, i : model.var[i] == model.aux_var[i])
logger("\nModel size before embedding model", export_log)
logger(f" #Variables: {model.nvariables()}", export_log)
logger(f" #Constraints: {model.nconstraints()}", export_log)
logger("\nStart embedding predictive models", export_log)
## EMPIRICAL CONSTRAINTS GRADIENT BOOSTED TREES ##
for target in set(list(user_constraint.keys()) + [objective['target']]):
# target price is not predicted, but indicated by the hw provider: it does not require any
# dedicated predictive model
if target == "price":
continue
# target quality does not depend on the hw: it requires a unique predictive model
elif target == "quality":
setattr(model, f'block_{target}', OmltBlock())
block = getattr(model, f'block_{target}')
gbt = pickle.load(open(mlmodel_files[f"{target}"], 'rb'))
scaler, input_bounds = get_scaling(algorithm=algorithm, target=target, hw=None)
# initial type defines type and length of model input variable
initial_type = [('input', FloatTensorType([None, ALG_PARAMS[algorithm]]))]
onx = convert_sklearn(gbt, initial_types=initial_type)
gbt_model = GradientBoostedTreeModel(onx, scaling_object=scaler, scaled_input_bounds=input_bounds)
formulation = GBTBigMFormulation(gbt_model)
block.build_formulation(formulation)
block.connect_input = pyo.Constraint(model.index_var, rule=lambda block, i : model.aux_var[i] == block.inputs[i])
block.connect_output = pyo.Constraint(expr=model.y[f'{target}'] == block.outputs[0])
# add_formulation_to_block(block, gbt_model, block.aux_var, [model.y[target]])
logger(f" Predictive model corresponding to ({target}) embedded", export_log)
# time and memory depend on both the hw and the algorithm configuration: each of them requires three
# dedicated predictive models
else:
for hw in HW:
setattr(model, f'block_{hw}_{target}', OmltBlock())
block = getattr(model, f'block_{hw}_{target}')
gbt = pickle.load(open(mlmodel_files[(f"{hw}", f"{target}")], 'rb'))
scaler, input_bounds = get_scaling(algorithm=algorithm, target=target, hw=hw)
# initial type defines type and length of model input variable
initial_type = [('input', FloatTensorType([None, ALG_PARAMS[algorithm]]))]
onx = convert_sklearn(gbt, initial_types=initial_type)
gbt_model = GradientBoostedTreeModel(onx, scaling_object=scaler, scaled_input_bounds=input_bounds)
formulation = GBTBigMFormulation(gbt_model)
block.build_formulation(formulation)
block.connect_input = pyo.Constraint(model.index_var, rule=lambda block, i : model.aux_var[i] == block.inputs[i])
block.connect_output = pyo.Constraint(expr=model.y[f'{hw}_{target}'] == block.outputs[0])
# add_formulation_to_block(block, gbt_model, block.aux_var, [model.y[f'{hw}_{target}']])
logger(f" Predictive model corresponding to ({hw}, {target}) embedded", export_log)
#print(model.block_pc_price)
logger("\nModel size after embedding empirical components", export_log)
logger(f" #Variables: {model.nvariables()}", export_log)
logger(f" #Constraints: {model.nconstraints()}", export_log)
logger("\nHardware prices:", export_log)
for hw in HW:
logger(f" {hw} = {price[hw]}", export_log)
# Handling non-estimated target (price) and robustness coefficients:
# 1.Equality constraints, fixing each price variable y_hw_price to the usage price of the corresponding hw,
# as required by the hw provider
if 'price' in list(user_constraint.keys()) + [objective['target']]:
model.index_constraint_price_hw = pyo.Set(initialize=[f"{hw}_price" for hw in HW])
model.constraint_price_hw = pyo.Constraint(model.index_constraint_price_hw, rule=lambda model, i: model.y[i] == price[i.split("_")[0]])
logger("\nUser requirements:", export_log)
for target in user_constraint:
logger(f" {target} {user_constraint[target]['type']} {user_constraint[target]['bound']}", export_log)
# 2. If no robustness is required, fix all coefficients to 0
if robust_coeff is None:
robust_coeff = {(hw, target) : 0 for hw in HW for target in user_constraint.keys() if target != 'quality'}
robust_coeff['quality'] = 0
logger("\nRoubstness Coefficients:", export_log)
for key in robust_coeff.keys():
logger(f" {key} : {robust_coeff[key]}", export_log)
# User-defined constraints, bounding the performance of the algorithm, as required by the user
model.user_constraint = pyo.ConstraintList()
for target in user_constraint.keys():
if target == 'quality':
if user_constraint[target]["type"] == "leq":
model.user_constraint.add(expr=model.y[target] <= user_constraint[f"{target}"]["bound"] - robust_coeff[(f'{target}')] )
elif user_constraint[target]["type"] == "geq":
model.user_constraint.add(expr=model.y[target] >= user_constraint[f"{target}"]["bound"] + robust_coeff[(f'{target}')])
elif user_constraint[target]["type"] == "eq":
model.user_constraint.add(expr=model.y[target] >= user_constraint[f"{target}"]["bound"] - robust_coeff[(f'{target}')])
model.user_constraint.add(expr=model.y[target] <= user_constraint[f"{target}"]["bound"] + robust_coeff[(f'{target}')])
else:
for hw in HW:
# choose M as the smallest value that
# trivially satisfies the constraint
M = model.y[f"{hw}_{target}"].ub + user_constraint[f"{target}"]["bound"] + robust_coeff[(f'{hw}',f'{target}')]
if user_constraint[target]["type"] == "leq":
model.user_constraint.add(
model.y[f"{hw}_{target}"] <= user_constraint[f"{target}"]["bound"] - robust_coeff[(f'{hw}',f'{target}')] + M*(1 - model.b[hw]))
elif user_constraint[target]["type"] == "geq":
model.user_constraint.add(
model.y[f"{hw}_{target}"] >= user_constraint[f"{target}"]["bound"] + robust_coeff[(f'{hw}',f'{target}')] - M*(1 - model.b[hw]))
elif user_constraint[target]["type"] == "eq":
model.user_constraint.add(
model.y[f"{hw}_{target}"] >= user_constraint[f"{target}"]["bound"] - robust_coeff[(f'{hw}',f'{target}')] - M*(1 - model.b[hw]))
model.user_constraint.add(
model.y[f"{hw}_{target}"] <= user_constraint[f"{target}"]["bound"] + robust_coeff[(f'{hw}',f'{target}')] + M*(1 - model.b[hw]))
end_time = time.time()
build_time = round(end_time - start_time, 2)
logger(f"\n==> Building process terminated in {build_time}s", export_log)
##### OBJECTIVE #####
logger("\nObjective", export_log)
logger(f" {objective['type']} {objective['target']}", export_log)
if objective["target"] == "quality":
if objective["type"] == "min":
model.objective = pyo.Objective(expr=model.y["quality"], sense=pyo.minimize)
else:
model.objective = pyo.Objective(expr=model.y["quality"], sense=pyo.maximize)
else:
if objective["type"] == "min":
model.objective = pyo.Objective(expr=sum(model.y[f"{hw}_{objective['target']}"] * model.b[hw] for hw in HW), sense=pyo.minimize)
else:
model.objective = pyo.Objective(expr=sum(model.y[f"{hw}_{objective['target']}"] * model.b[hw] for hw in HW), sense=pyo.maximize)
solver = pyo.SolverFactory('cplex') # cplex needs compiled extension
##### SOLVE #####
logger("\n==============> Solving", export_log)
start_time = time.time()
results = solver.solve(model, symbolic_solver_labels=True, tee=True)
end_time = time.time()
solve_time = round(end_time - start_time, 2)
logger(f"\n==> Solving process terminated in {solve_time}s due to {results.solver.termination_condition}", export_log)
if results.solver.termination_condition == TerminationCondition.optimal.name:
solution = dict(
{"objective" : (objective["type"], objective["target"])},
**{target : (user_constraint[target]["type"], user_constraint[target]["bound"]) if target in user_constraint.keys() else "-" for target in TARGET},
**{v.name : v.value for v in model.component_data_objects(pyo.Var) if v.name in DECLARED_VARS },
**{"#variables" : model.nvariables()},
**{"#constraints" : model.nconstraints()},
**{"build_time" : build_time, "solve_time" : solve_time}
)
else:
print('The following termination condition was encountered: ', results.solver.termination_condition)
solution = dict(
{"objective" : (objective["type"], objective["target"])},
**{target : (user_constraint[target]["type"], user_constraint[target]["bound"]) if target in user_constraint.keys() else "-" for target in TARGET},
**{v.name : None for v in model.component_data_objects(pyo.Var) if v.name in DECLARED_VARS},
**{"#variables" : model.nvariables()},
**{"#constraints" : model.nconstraints()},
**{"build_time" : build_time, "solve_time" : solve_time}
)
model.write("sol.lp")
return solution, model
if __name__ == "__main__":
####### INPUT #######
parser = argparse.ArgumentParser(prog = 'HADA', description = 'An optimization engine for hardware dimensioning and algorithm configuration, designed as the vertical matchmaking\nlayer of the European AI on-demand platform.')
parser.formatter_class = lambda prog: argparse.RawTextHelpFormatter(prog, max_help_position=55)
parser.add_argument('--data_folder', type = str, default = "datasets", required = False,
help = 'folder with data for bounds/robustness coefficients (def: datasets)')
parser.add_argument('--models_folder', type = str, default = "GBTs", required = False,
help = 'folder with predictive models (def: GBTs)')
parser.add_argument('--algorithm', type = str, required = True,
help = 'algorithm to execute', choices = ['saxpy', 'convolution', 'correlation', 'fwt'])
# hw prices
parser.add_argument('--price_pc', type = float, default = 7, required = False,
help = 'usage price of pc hardware (def: 7)')
parser.add_argument('--price_g100', type = float, default = 22, required = False,
help = 'usage price of g100 hardware (def: 22)')
parser.add_argument('--price_vm', type = float, default = 38, required = False,
help = 'usage price of vm hardware (def: 38)')
# objective
parser.add_argument('--objective', type = str, required = True, nargs = 2,
help = 'objective to optimize', metavar = ('{min,max}', 'TARGET'))
# target bounds
parser.add_argument('--constraint_time', type = str, required = False, nargs = 2,
help = 'user constraint for time', metavar = ('{leq,eq,geq}', 'BOUND'))
parser.add_argument('--constraint_quality', type = str, required = False, nargs = 2,
help = 'user constraint for quality', metavar = ('{leq,eq,geq}', 'BOUND'))
parser.add_argument('--constraint_memory', type = str, required = False, nargs = 2,
help = 'user constraint for memory', metavar = ('{leq,eq,geq}', 'BOUND'))
parser.add_argument('--constraint_price', type = str, required = False, nargs = 2,
help = 'user constraint for price', metavar = ('{leq,eq,geq}', 'BOUND'))
# robustness factor
parser.add_argument('--robust_factor', type = float, default = None, required = False,
help = 'confidence margin to introduce into the user constraints (def: None)')
# model/results storing option
parser.add_argument('--export_model', default=False, action='store_true',
help = 'when given, the final HADA model is exported to a .lp file')
parser.add_argument('--export_results', default=False, action='store_true',
help = 'when given, the obtained solution is exported to a .csv file')
parser.add_argument('--export_log', default=False, action='store_true',
help = 'when given, the log is exported to a .log file')
parser.add_argument('--train', default=False, action='store_true',
help = 'when given, the GBTs are retrained with the given max_depth, estimators and loss hyperparameters')
parser.add_argument('--max_depth', type=int, default=10, required=False,
help = 'max_depth hyperparameter for GBTs')
parser.add_argument('--estimators', type=int, default=1, required=False,
help = 'number of estimators for GBTs')
args = parser.parse_args()
logger("=================================================================", args.export_log)
logger(time.ctime(), args.export_log)
logger(f"Running HADA for {args.algorithm} algorithm", args.export_log)
if args.train:
train_quality_GBT("datasets", estimators=args.estimators, max_depth=args.max_depth)
train_targets_GBT("datasets", estimators=args.estimators, max_depth=args.max_depth)
##### TREES ######
# load the trained empirical model to embed into HADA
mlmodel_files = {(f"{hw}", f"{target}") : f"{args.models_folder}/{args.algorithm}_{hw}_{target}_GradientBoostingRegressor_{args.max_depth}_{args.estimators}" for hw in HW for target in TARGET if target != "quality"}
mlmodel_files["quality"] = f"{args.models_folder}/{args.algorithm}_quality_GradientBoostingRegressor_{args.max_depth}_{args.estimators}"
###### USER DATA #####
price = {"pc" : args.price_pc, "vm" : args.price_vm, "g100" : args.price_g100}
user_constraint = {
"price" : args.constraint_price,
"time" : args.constraint_time,
"quality" : args.constraint_quality,
"memory" : args.constraint_memory
}
user_constraint = {target : {"type" : constraint[0], "bound" : float(constraint[1])}
for target, constraint in user_constraint.items() if constraint is not None}
objective = {"type" : args.objective[0], "target" : args.objective[1]}
###### DATA #####
# extract, from the training datasets of the empirical models,
# 1. upper/lower bound for each declarative variable
# 2. robustness coefficient for each predictive model
var_bounds = extract_var_bounds(
data_folder = args.data_folder,
algorithm = args.algorithm,
price = price)
robust_coeff = extract_robust_coeff(
robust_factor = args.robust_factor,
data_folder = args.data_folder,
algorithm = args.algorithm,
mlmodel_files = mlmodel_files,
TARGET = list(user_constraint.keys()) + [objective['target']])
# Build-and-Solve
start_time = time.time()
solution, model = HADA(
algorithm = args.algorithm,
objective = objective,
user_constraint = user_constraint,
price = price,
var_bounds = var_bounds,
mlmodel_files = mlmodel_files,
robust_coeff = robust_coeff,
export_log = args.export_log
)
end_time = time.time()
solution["robust_factor"] = args.robust_factor
# Print optimal solution
if solution["b[pc]"] is not None:
logger("\n==============> Optimal solution found:", args.export_log)
log_solution = pd.DataFrame({f"var[{i}]" : [None if solution[f"var[{i}]"] is None else round(solution[f"var[{i}]"])] for i in range(ALG_PARAMS[args.algorithm])})
log_solution["hw"] = [round(solution["b[pc]"])*1 + round(solution["b[g100]"])*2 + round(solution["b[vm]"]*3)]
log_solution["hw"] = log_solution["hw"].replace({1 : "pc", 2 : "g100", 3 : "vm"})
for target in list(user_constraint.keys()) + [objective["target"]]:
if target == "price":
log_solution[target] = [round(solution[f"y[{log_solution.loc[0, 'hw']}_{target}]"], 2)]
elif target == "quality":
log_solution[f"pred {target}"] = [round(solution[f"y[{target}]"], 2)]
else:
log_solution[f"pred {target}"] = [round(solution[f"y[{log_solution.loc[0, 'hw']}_{target}]"], 2)]
log_table = PrettyTable()
log_table.field_names = log_solution
log_table.add_row(log_solution.iloc[0, :])
logger(f"\n{log_table}", args.export_log)
else:
logger("\n==============> No solution found", args.export_log)
if args.export_results:
# create fixed column names for csv
# otherwise different executions will load wrong
# target names
fieldnames = ['objective']
fieldnames += [target for target in TARGET]
fieldnames += [f"b[{hw}]" for hw in HW]
fieldnames += [f"y[{hw}_{target}]" for hw in HW for target in TARGET if target != "quality"] + ["y[quality]"]
fieldnames += [f"var[{i}]" for i in range(ALG_PARAMS[args.algorithm])]
fieldnames += ["#variables", "#constraints", "build_time", "solve_time", "robust_factor"]
filename = f"./{args.algorithm}_depth{args.max_depth}_est{args.estimators}_results.csv"
if not os.path.exists(filename):
# results = pd.DataFrame({k : [] for k in solution.keys()})
with open(filename, 'w') as csvfile:
writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
writer.writeheader()
writer.writerow(solution)
# results.to_csv(f"./{args.algorithm}_results.csv", index = False)
else:
with open(filename,'a') as csvfile:
writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
#writer=csv.writer(f)
writer.writerow(solution)
logger(f"\nResults exported into {filename}", args.export_log)
if args.export_model:
logger("\nExporting model", args.export_log)
model.export_as_lp("hada.lp")
logger("Model exported into hada.lp", args.export_log)