Q&A前提・実現したいこと
制約付き非線形関数の最小化問題で制約条件の係数の値を色々変えながら
scipy.optimize.minimizeでそれぞれの値に対応する最小値を全て取得して結果を配列に格納する
該当のソースコード
python
1import numpy as np 2from scipy.optimize import minimize 3 4def objective_function(x): 5 return - x[0]**0.5 * x[1]**0.5 6 7def constraint(x,a,b): 8 x,a = np.array(x),np.array(a) 9 return b - x@a 10 11reps = 100 #繰り返し回数 12 13a_init = [50,200] #制約条件係数の基準値 14a_shock = np.array([np.random.rand(reps)*50+50,np.zeros(reps)]).T #係数をランダムに変更 15a = a_init + a_shock 16b = 10000 17 18cons = lambda x,i: constraint(x,a[i],b) 19cons_list = [lambda x: cons(x,i) for i in range(len(a))] 20bnds = ((0,None),(0,None)) 21opts = lambda i: minimize(objective_function, #目的関数 22 (15,5), #初期値 23 method='SLSQP', 24 bounds=bnds, 25 constraints={'type':'eq', 'fun':cons_list[i]}) 26#以下を実行 27print(opts(0).x) 28# array([33.62294395, 25.00001409]) 29 30#異なるインデックスに対して再度実行すると別の値が返ってくる 31print(opts(1).x) 32# array([35.37724898, 25.0000091 ]) 33 34#しかし以下のように配列化すると全て同じ結果が返ってくる 35results = [opts(i).x for i in range(len(a))] 36print(results) 37# [array([37.31991237, 24.99999784]), array([37.31991237, 24.99999784]), ... 38 39#一度の実行で複数のインデックスについて計算すると同じ値が返ってくる 40print(f"インデックス1: {opts(1).x}\nインデックス2: {opts(2).x}") 41# インデックス1: [44.29645281 25.00000902] 42# インデックス2: [44.29645281 25.00000902]
試したこと
〇普通にfor loopしたり、リスト内記法で結果を取得すると全て同じ値になってしまう
〇lambda関数化して1つずつ結果を取り出すと問題なく毎回異なる結果が返ってくる
〇↑をforで回して結果を配列に格納してもすべて同じ値になってしまう
補足情報（FW/ツールのバージョンなど）
実行環境はGoogle colaboratoryです
回答3件
0
解決しているので蛇足かも知れませんが、ラムダ式の内包表現に問題ありです。
python
1def func(x, a): 2 print(f"x={x},a={a},id(a)={id(a)}") 3 4 5a = list(range(10)) 6print("a=", a) 7 8lfunc = lambda x, i: func(x, a[i]) 9lfuncs = [lambda x: lfunc(x, i) for i in range(len(a))] 10 11lfuncs[0](123) 12lfuncs[1](456) 13lfuncs[2](789)
結果
python
1a= [0, 1, 2, 3, 4, 5, 6, 7, 8, 9] 2x=123,a=9,id(a)=4452857216 3x=456,a=9,id(a)=4452857216 4x=789,a=9,id(a)=4452857216
投稿2020/07/03 03:27
編集2020/07/03 03:29
総合スコア1615
0
自己解決
問題はどうやらcons_listだったようです。
【該当部分】
python
1cons_list = [lambda x: cons(x,i) for i in range(len(a))]
ここでリスト内記法を用いると中身の関数が全て同じになってしまうようです。
なので、cons_listは作らずにoptsを以下のようにすると異なる結果を一度に取得できました。
python
1import numpy as np 2from scipy.optimize import minimize 3 4def objective_function(x): 5 return - x[0]**0.5 * x[1]**0.5 6 7def constraint(x,a,b): 8 x,a = np.array(x),np.array(a) 9 return b - x@a 10 11reps = 100 #繰り返し回数 12 13a_init = [50,200] #制約条件係数の基準値 14a_shock = np.array([np.random.rand(reps)*50+50,np.zeros(reps)]).T #係数をランダムに変更 15a = a_init + a_shock 16b = 10000 17 18cons = lambda x,i: constraint(x,a[i],b) 19bnds = ((0,None),(0,None)) 20 21opts = lambda i: minimize(objective_function, #目的関数 22 (15,5), #初期値 23 method='SLSQP', 24 bounds=bnds, 25 constraints={'type':'eq', 'fun':lambda x:cons(x,i)}) 26 27res = [opts(i).x for i in range(reps)] 28res
実行結果
python
1#毎回結果が異なっている 2[array([37.59090689, 25.00025669]), array([40.62532566, 24.99999821]), array([45.94143763, 24.99998405]), array([38.30967904, 24.99999733]), ...
投稿2020/07/03 00:55
総合スコア2
0
ご質問のコードを何度実行しても再現できませんでした(全て同じ値が出力されました)。
(ある特定のa_shockが原因か、あるいは、モジュールのバージョンが原因かわかりません)
少なくともprint(opts(0).x)とprint(opts(1).x)が異なることはありませんでした。
本当にご質問のような結果が得られたのですか？
text
1[41.14691214 24.99999917] 2[41.14691214 24.99999917] 3[array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917]), array([41.14691214, 24.99999917])] 4インデックス1: [41.14691214 24.99999917] 5インデックス2: [41.14691214 24.99999917]
この結果を再現したa_shockの値
text
1array([[86.94631736, 0. ], 2 [88.24059994, 0. ], 3 [63.88071978, 0. ], 4 [97.57583615, 0. ], 5 [59.1515068 , 0. ], 6 [76.57500374, 0. ], 7 [62.57463254, 0. ], 8 [71.99813083, 0. ], 9 [79.11855668, 0. ], 10 [84.37097279, 0. ], 11 [56.33228482, 0. ], 12 [76.90794869, 0. ], 13 [88.78803925, 0. ], 14 [72.86546644, 0. ], 15 [64.55730109, 0. ], 16 [87.66662926, 0. ], 17 [68.26486145, 0. ], 18 [89.0884346 , 0. ], 19 [51.9560971 , 0. ], 20 [91.22909041, 0. ], 21 [75.41843306, 0. ], 22 [91.10976738, 0. ], 23 [73.51168527, 0. ], 24 [89.40328009, 0. ], 25 [53.81101371, 0. ], 26 [61.98393568, 0. ], 27 [54.71236022, 0. ], 28 [67.17937382, 0. ], 29 [98.55481394, 0. ], 30 [77.88477456, 0. ], 31 [73.64123646, 0. ], 32 [94.04620222, 0. ], 33 [58.52606914, 0. ], 34 [81.7636329 , 0. ], 35 [69.39601077, 0. ], 36 [72.49074165, 0. ], 37 [77.33081577, 0. ], 38 [77.02725975, 0. ], 39 [59.50171833, 0. ], 40 [95.2699818 , 0. ], 41 [97.16985503, 0. ], 42 [83.2670286 , 0. ], 43 [52.5908398 , 0. ], 44 [56.48330684, 0. ], 45 [87.88075307, 0. ], 46 [63.01896622, 0. ], 47 [89.89313833, 0. ], 48 [51.57691706, 0. ], 49 [81.922707 , 0. ], 50 [94.3653957 , 0. ], 51 [85.20211311, 0. ], 52 [78.60473673, 0. ], 53 [60.04666014, 0. ], 54 [54.09208572, 0. ], 55 [72.95415372, 0. ], 56 [75.09743132, 0. ], 57 [60.1501105 , 0. ], 58 [68.80185706, 0. ], 59 [83.84797023, 0. ], 60 [98.53367147, 0. ], 61 [65.52482817, 0. ], 62 [50.90285463, 0. ], 63 [82.66954479, 0. ], 64 [88.0250928 , 0. ], 65 [60.12744854, 0. ], 66 [95.11834392, 0. ], 67 [98.78607007, 0. ], 68 [98.39624051, 0. ], 69 [94.87834786, 0. ], 70 [76.22672687, 0. ], 71 [60.67453039, 0. ], 72 [99.30529285, 0. ], 73 [67.59547283, 0. ], 74 [70.21665278, 0. ], 75 [58.52796051, 0. ], 76 [82.91705426, 0. ], 77 [99.79190275, 0. ], 78 [94.23850625, 0. ], 79 [96.80085235, 0. ], 80 [88.94066684, 0. ], 81 [88.09091175, 0. ], 82 [52.63167066, 0. ], 83 [82.86003197, 0. ], 84 [91.62778404, 0. ], 85 [98.19051657, 0. ], 86 [69.39058311, 0. ], 87 [97.77421829, 0. ], 88 [83.57513457, 0. ], 89 [54.43381656, 0. ], 90 [54.03169961, 0. ], 91 [54.59933567, 0. ], 92 [78.88361243, 0. ], 93 [82.9944889 , 0. ], 94 [55.94026688, 0. ], 95 [56.15304813, 0. ], 96 [63.84645291, 0. ], 97 [56.79068478, 0. ], 98 [97.46527693, 0. ], 99 [86.68915797, 0. ], 100 [64.58626962, 0. ]])
投稿2020/07/02 17:33
編集2020/07/03 00:55
総合スコア698
あなたの回答
tips
プレビュー
バッドをするには、ログインかつ
こちらの条件を満たす必要があります。