1"Scripts for generating, solving and sweeping programs" 

2from time import time 

3import numpy as np 

4from ad import adnumber 

5from ..nomials import parse_subs 

6from ..solution_array import SolutionArray 

7from ..keydict import KeyDict 

8from ..small_scripts import maybe_flatten 

9from ..small_classes import FixedScalar 

10from ..exceptions import Infeasible 

11from ..globals import SignomialsEnabled 

12 

13 

14def evaluate_linked(constants, linked): 

15 "Evaluates the values and gradients of linked variables." 

16 kdc = KeyDict({k: adnumber(maybe_flatten(v), k) 

17 for k, v in constants.items()}) 

18 kdc_plain = None 

19 array_calulated = {} 

20 for key in constants: # remove gradients from constants 

21 if key.gradients: 

22 del key.descr["gradients"] 

23 for v, f in linked.items(): 

24 try: 

25 if v.veckey and v.veckey.vecfn: 

26 if v.veckey not in array_calulated: 

27 with SignomialsEnabled(): # to allow use of gpkit.units 

28 vecout = v.veckey.vecfn(kdc) 

29 if not hasattr(vecout, "shape"): 

30 vecout = np.array(vecout) 

31 array_calulated[v.veckey] = vecout 

32 if (any(vecout != 0) and v.veckey.units 

33 and not hasattr(vecout, "units")): 

34 print("Warning: linked function for %s did not return" 

35 " a united value. Modifying it to do so (e.g. by" 

36 " using `()` instead of `[]` to access variables)" 

37 " would reduce the risk of errors." % v.veckey) 

38 out = array_calulated[v.veckey][v.idx] 

39 else: 

40 with SignomialsEnabled(): # to allow use of gpkit.units 

41 out = f(kdc) 

42 if isinstance(out, FixedScalar): # to allow use of gpkit.units 

43 out = out.value 

44 if hasattr(out, "units"): 

45 out = out.to(v.units or "dimensionless").magnitude 

46 elif out != 0 and v.units and not v.veckey: 

47 print("Warning: linked function for %s did not return" 

48 " a united value. Modifying it to do so (e.g. by" 

49 " using `()` instead of `[]` to access variables)" 

50 " would reduce the risk of errors." % v) 

51 if not hasattr(out, "x"): 

52 constants[v] = out 

53 continue # a new fixed variable, not a calculated one 

54 constants[v] = out.x 

55 gradients = {adn.tag: 

56 grad for adn, grad in out.d().items() if adn.tag} 

57 if gradients: 

58 v.descr["gradients"] = gradients 

59 except Exception as exception: # pylint: disable=broad-except 

60 from .. import settings 

61 if settings.get("ad_errors_raise", None): 

62 raise 

63 print("Warning: skipped auto-differentiation of linked variable" 

64 " %s because %s was raised. Set `gpkit.settings" 

65 "[\"ad_errors_raise\"] = True` to raise such Exceptions" 

66 " directly.\n" % (v, repr(exception))) 

67 if kdc_plain is None: 

68 kdc_plain = KeyDict(constants) 

69 constants[v] = f(kdc_plain) 

70 v.descr.pop("gradients", None) 

71 

72 

73def progify(program, return_attr=None): 

74 """Generates function that returns a program() and optionally an attribute. 

75 

76 Arguments 

77 --------- 

78 program: NomialData 

79 Class to return, e.g. GeometricProgram or SequentialGeometricProgram 

80 return_attr: string 

81 attribute to return in addition to the program 

82 """ 

83 def programfn(self, constants=None, **initargs): 

84 "Return program version of self" 

85 if not constants: 

86 constants, _, linked = parse_subs(self.varkeys, self.substitutions) 

87 if linked: 

88 evaluate_linked(constants, linked) 

89 prog = program(self.cost, self, constants, **initargs) 

90 prog.model = self # NOTE SIDE EFFECTS 

91 if return_attr: 

92 return prog, getattr(prog, return_attr) 

93 return prog 

94 return programfn 

95 

96 

97def solvify(genfunction): 

98 "Returns function for making/solving/sweeping a program." 

99 def solvefn(self, solver=None, *, verbosity=1, skipsweepfailures=False, 

100 **solveargs): 

101 """Forms a mathematical program and attempts to solve it. 

102 

103 Arguments 

104 --------- 

105 solver : string or function (default None) 

106 If None, uses the default solver found in installation. 

107 verbosity : int (default 1) 

108 If greater than 0 prints runtime messages. 

109 Is decremented by one and then passed to programs. 

110 skipsweepfailures : bool (default False) 

111 If True, when a solve errors during a sweep, skip it. 

112 **solveargs : Passed to solve() call 

113 

114 Returns 

115 ------- 

116 sol : SolutionArray 

117 See the SolutionArray documentation for details. 

118 

119 Raises 

120 ------ 

121 ValueError if the program is invalid. 

122 RuntimeWarning if an error occurs in solving or parsing the solution. 

123 """ 

124 constants, sweep, linked = parse_subs(self.varkeys, self.substitutions) 

125 solution = SolutionArray() 

126 solution.modelstr = str(self) 

127 

128 # NOTE SIDE EFFECTS: self.program and self.solution set below 

129 if sweep: 

130 run_sweep(genfunction, self, solution, skipsweepfailures, 

131 constants, sweep, linked, solver, verbosity, **solveargs) 

132 else: 

133 self.program, progsolve = genfunction(self) 

134 result = progsolve(solver, verbosity=verbosity, **solveargs) 

135 if solveargs.get("process_result", True): 

136 self.process_result(result) 

137 solution.append(result) 

138 solution.to_arrays() 

139 self.solution = solution 

140 return solution 

141 return solvefn 

142 

143 

144# pylint: disable=too-many-locals,too-many-arguments,too-many-branches 

145def run_sweep(genfunction, self, solution, skipsweepfailures, 

146 constants, sweep, linked, solver, verbosity, **solveargs): 

147 "Runs through a sweep." 

148 # sort sweeps by the eqstr of their varkey 

149 sweepvars, sweepvals = zip(*sorted(list(sweep.items()), 

150 key=lambda vkval: vkval[0].eqstr)) 

151 if len(sweep) == 1: 

152 sweep_grids = np.array(list(sweepvals)) 

153 else: 

154 sweep_grids = np.meshgrid(*list(sweepvals)) 

155 

156 N_passes = sweep_grids[0].size 

157 sweep_vects = {var: grid.reshape(N_passes) 

158 for (var, grid) in zip(sweepvars, sweep_grids)} 

159 

160 if verbosity > 0: 

161 print("Sweeping with %i solves:" % N_passes) 

162 tic = time() 

163 

164 self.program = [] 

165 last_error = None 

166 for i in range(N_passes): 

167 constants.update({var: sweep_vect[i] 

168 for (var, sweep_vect) in sweep_vects.items()}) 

169 if linked: 

170 evaluate_linked(constants, linked) 

171 program, solvefn = genfunction(self, constants) 

172 self.program.append(program) # NOTE: SIDE EFFECTS 

173 try: 

174 if verbosity > 1: 

175 print("\nSolve %i:" % i) 

176 result = solvefn(solver, verbosity=verbosity-1, **solveargs) 

177 if solveargs.get("process_result", True): 

178 self.process_result(result) 

179 solution.append(result) 

180 except Infeasible as e: 

181 last_error = e 

182 if not skipsweepfailures: 

183 raise RuntimeWarning( 

184 "Solve %i was infeasible; progress saved to m.program." 

185 " To continue sweeping after failures, solve with" 

186 " skipsweepfailures=True." % i) from e 

187 if verbosity > 0: 

188 print("Solve %i was %s." % (i, e.__class__.__name__)) 

189 if not solution: 

190 raise RuntimeWarning("All solves were infeasible.") from last_error 

191 

192 solution["sweepvariables"] = KeyDict() 

193 ksweep = KeyDict(sweep) 

194 for var, val in list(solution["constants"].items()): 

195 if var in ksweep: 

196 solution["sweepvariables"][var] = val 

197 del solution["constants"][var] 

198 elif linked: # if any variables are linked, we check all of them 

199 if hasattr(val[0], "shape"): 

200 differences = ((l != val[0]).any() for l in val[1:]) 

201 else: 

202 differences = (l != val[0] for l in val[1:]) 

203 if not any(differences): 

204 solution["constants"][var] = [val[0]] 

205 else: 

206 solution["constants"][var] = [val[0]] 

207 

208 if verbosity > 0: 

209 soltime = time() - tic 

210 print("Sweeping took %.3g seconds." % (soltime,))