anzu_chan_no_op.py

# 1クラス認識しかできないよ
import numpy as np

def sigmoid(x):
    return 1 / (1 + np.exp(-x))
def d_sigmoid(x):
    return sigmoid(x) * (1.0 - sigmoid(x))

# 各層の素子数(項数)定義 + 1はバイアス項の分
input_symbol = 3 + 1
hidden_symbol = 15 + 1
output_symbol = 1

# 学習回数の指定(データセットの学習を何週分するか)
max_epoch = 10
# 学習係数
p = 0.05

error = []

#196人分
#年齢,   身長,  体重, B, W, H
idol = [
[23,	161,	43,	82,	56,	85],
[18,	163,	47,	84,	58,	85],
[22,	160,	51,	92,	58,	90],
[18,	158,	42,	81,	57,	80],
[11,	140,	36,	75,	55,	78],
[16,	154,	55,	92,	59,	88],
[19,	156,	43,	85,	57,	85],
[16,	160,	48,	88,	59,	84],
[14,	151,	41,	78,	55,	77],
[15,	165,	43,	80,	54,	80],
[17,	158,	46,	83,	56,	82],
[17,	161,	46,	85,	57,	84],
[20,	157,	43,	83,	57,	82],
[19,	155,	44,	78,	57,	80],
[16,	156,	41,	78,	55,	77],
[15,	154,	43,	81,	58,	80],
[14,	148,	39,	75,	53,	74],
[21,	160,	44,	86,	56,	81],
[18,	161,	43,	83,	57,	82],
[9,     128,    29, 61, 57, 67],
[19,	165,	44,	81,	56,	80],
[16,	153,	41,	81,	56,	79],
[17,	163,	48,	85,	60,	88],
[14,	156,	42,	76,	55,	79],
[13,	152,	42,	78,	57,	80],
[19,	172,	49,	86,	58,	85],
[19,	161,	46,	80,	57,	80],
[22,	160,	45,	84,	56,	80],
[17,	162,	58,	92,	65,	93],
[16,	170,	56,	105,64,	92],
[15,	157,	41,	83,	55,	82],
[21,	159,	45,	89,	57,	87],
[17,	155,	42,	84,	56,	83],
[20,	156,	44,	81,	58,	83],
[15,	153,	40,	78,	55,	80],
[16,	153,	43,	79,	55,	80],
[16,	153,	42,	79,	57,	80],
[19,	156,	47,	83,	57,	81],
[13,	164,	40,	70,	53,	74],
[28,	152,	47,	92,	58,	84],
[18,	156,	42,	79,	56,	80],
[17,	154,	44,	83,	58,	81],
[28,	159,	44,	87,	57,	85],
[14,	156,	41,	81,	57,	80],
[16,	152,	41,	83,	56,	80],
[17,	159,	44,	75,	57,	78],
[16,	162,	41,	72,	55,	78],
[19,	162,	46,	89,	59,	85],
[17,	158,	43,	75,	57,	79],
[15,	151,	38,	78,	56,	78],
[15,	156,	43,	82,	57,	82],
[25,	172,	50,	88,	60,	89],
[18,	159,	41,	82,	57,	83],
[15,	168,	49,	83,	56,	85],
[19,	160,	43,	86,	56,	86],
[18,	164,	45,	83,	55,	82],
[20,	148,	40,	77,	54,	78],
[16,	154,	41,	78,	54,	81],
[20,	166,	45,	80,	55,	82],
[15,	161,	44,	77,	54,	78],
[20,	163,	45,	86,	57,	86],
[20,	157,	45,	83,	59,	85],
[12,	140,	35,	72,	54,	77],
[14,	142,	37,	74,	52,	75],
[13,	148,	41,	75,	50,	77],
[15,	148,	42,	78,	56,	80],
[17,	155,	42,	82,	59,	86],
[19,	165,	48,	85,	59,	88],
[19,	164,	45,	84,	56,	82],
[17,	156,	46,	87,	57,	85],
[20,	157,	46,	85,	57,	82],
[15,	153,	43,	76,	58,	78],
[17,	162,	46,	91,	56,	86],
[19,	162,	45,	84,	54,	81],
[16,	153,	40,	78,	54,	80],
[12,	145,	39,	72,	53,	75],
[11,	139,	33,	73,	49,	73],
[10,	137,	30,	63,	47,	65],
[26,	166,	47,	87,	57,	87],
[21,	168,	46,	83,	55,	85],
[13,	147,	38,	76,	55,	79],
[18,	163,	45,	82,	56,	81],
[18,	169,	49,	90,	62,	92],
[27,	171,	49,	93,	58,	88],
[15,	165,	44,	80,	56,	81],
[17,	159,	45,	83,	59,	87],
[13,	145,	39,	73,	53,	75],
[13,	156,	42,	77,	53,	76],
[13,	142,	34,	65,	50,	70],
[17,	162,	43,	80,	56,	82],
[12,	149,	36,	72,	54,	75],
[18,	162,	45,	88,	58,	86],
[19,	158,	45,	82,	57,	83],
[14,	155,	43,	78,	55,	80],
[19,	165,	48,	85,	58,	83],
[21,	156,	45,	81,	55,	81],
[25,	160,	46,	83,	60,	89],
[25,	171,	49,	81,	57,	83],
[14,	145,	37,	74,	54,	78],
[31,	167,	51,	91,	62,	90],
[20,	160,	43,	88,	57,	88],
[24,	168,	48,	87,	55,	86],
[16,	155,	42,	74,	60,	79],
[17,	152,	41,	80,	55,	81],
[12,	141,	34,	68,	52,	67],
[18,	153,	44,	86,	56,	82],
[15,	156,	42,	85,	54,	83],
[23,	167,	45,	82,	57,	83],
[18,	161,	46,	86,	58,	88],
[23,	160,	45,	84,	56,	85],
[17,	155,	43,	80,	57,	83],
[18,	149,	40,	77,	57,	81],
[16,	161,	45,	83,	56,	85],
[14,	155,	43,	84,	55,	86],
[22,	160,	44,	80,	57,	82],
[13,	150,	40,	72,	51,	73],
[14,	140,	41,	79,	58,	80],
[17,	158,	45,	82,	56,	80],
[16,	155,	55,	88,	60,	86],
[19,	172,	49,	86,	59,	83],
[19,	165,	45,	82,	55,	85],
[14,	154,	42,	75,	55,	78],
[18,	157,	43,	81,	55,	79],
[15,	157,	46,	84,	57,	85],
[17,	155,	42,	81,	56,	81],
[25,	169,	48,	78,	57,	80],
[22,	168,	50,	92,	58,	85],
[15,	154,	42,	78,	55,	80],
[14,	147,	39,	75,	54,	77],
[17,	162,	43,	86,	55,	84],
[20,	163,	46,	86,	59,	85],
[31,	167,	43,	84,	54,	83],
[17,	148,	40,	80,	60,	82],
[20,	161,	44,	80,	57,	80],
[27,	167,	48,	92,	56,	84],
[10,	132,	28,	64,	56,	70],
[19,	163,	45,	78,	57,	83],
[18,	154,	41,	77,	55,	80],
[16,	161,	43,	80,	55,	84],
[13,	158,	42,	78,	55,	77],
[13,	158,	42,	78,	55,	77],
[17,	165,	45,	80,	59,	83],
[24,	158,	46,	90,	58,	81],
[26,	163,	45,	83,	56,	84],
[16,	155,	42,	83,	55,	81],
[15,	161,	45,	86,	55,	83],
[15,	142,	35,	73,	53,	75],
[16,	157,	44,	81,	58,	80],
[15,	161,	46,	84,	58,	87],
[15,	152,	45,	85,	55,	81],
[19,	162,	46,	86,	57,	91],
[28,	166,	45,	82,	56,	83],
[15,	161,	45,	78,	54,	81],
[18,	160,	47,	90,	56,	86],
[18,	156,	43,	83,	58,	86],
[22,	165,	48,	92,	58,	85],
[21,	161,	44,	81,	56,	81],
[12,	143,	38,	71,	58,	73],
[20,	160,	46,	84,	57,	85],
[22,	165,	46,	85,	57,	83],
[21,	168,	48,	91,	59,	86],
[23,	155,	43,	82,	55,	80],
[18,	164,	47,	80,	54,	81],
[15,	155,	42,	81,	56,	82],
[15,	153,	40,	77,	54,	79],
[26,	165,	46,	85,	60,	85],
[17,	153,	52,	90,	65,	89],
[19,	164,	46,	83,	57,	85],
[14,	149,	39,	75,	56,	80],
[18,	163,	53,	95,	60,	87],
[14,	151,	41,	73,	56,	75],
[13,	146,	37,	74,	53,	76],
[15,	145,	38,	75,	55,	77],
[11,	150,	40,	70,	58,	72],
[21,	156,	45,	77,	54,	76],
[13,	150,	37,	82,	56,	86],
[15,	145,	40,	80,	55,	78],
[14,	149,	38,	73,	55,	76],
[17,	182,	60,	91,	64,	86],
[15,	152,	41,	82,	58,	84],
[18,	160,	45,	85,	56,	83],
[14,	150,	41,	81,	56,	80],
[23,	158,	47,	85,	58,	86],
[14,	144,	33,	75,	54,	77],
[21,	165,	51,	92,	60,	85],
[12,	140,	37,	74,	55,	78],
[11,	130,	28,	62,	50,	65],
[9 ,    127,    31, 60, 55, 65],
[17,	166,	43,	86,	60,	85],
[16,	151,	40,	73,	53,	73],
[16,	150,	40,	75,	54,	78],
[9,     132,    32, 65, 51, 70],
[19,	157,	45,	77,	56,	82],
[17,	156,	45,	82,	57,	83],
[16,	145,	38,	72,	53,	75],
[26,	168,	49,	81,	60,	86],
]

# 係数行列の初期化, 0-1の範囲
# 引数は 列,行のサイズ
W1 = np.random.random_sample((input_symbol, hidden_symbol - 1))
W2 = np.random.random_sample((hidden_symbol, output_symbol))

# ネットワークの学習
for epoch in range(0, max_epoch):
    err = 0
    
    for i in range(0, 195):
        # 学習データの入力
        X1 = np.array([1, 
        idol[i][0] / 31,
        idol[i][1] /182,
        idol[i][2] / 60])
        # 教師信号(出力してほしい値)
        S = np.array(idol[i][3])

        # 中間層の計算
        # 入力と係数のドット積
        X2 = np.dot(X1, W1)
        # 結果に活性化関数をかける
        O2 = sigmoid(X2)
        # 出力層の計算
        # 行列が逆になっているので転置する
        O2 = np.r_[1, O2]
        # 中間層の出力と係数のドット積
        X3 = np.dot(O2, W2)

        # back proapgation
        # 誤差表示用
        err += abs(X3 - S)
        # δoutの計算
        D2 = X3 - S
        # δhiddenの計算
        D1 = d_sigmoid(np.r_[1, X2]) * D2 * W2.T

        # 重みの修正
        # X1の転置
        X1 = X1[np.newaxis, :]
        # δoutのバイアス項はδhiddenに伝播しないので除外する
        D1 = D1[0][1:]

        W1 = W1 - p * (X1.T * D1)
        
        # D2, O2の転置
        D2 = D2[:, np.newaxis]
        O2 = O2[np.newaxis, :]
        
        W2 = W2 - p * (O2.T * D2)

    print(epoch, ", err = ", err)

# 訓練済みネットワークへの入力
X1 = np.array([1,
17 / 31,
139 / 182,
30 / 60])

# 入力と係数のドット積
X2 = np.dot(X1, W1)
# 結果に活性化関数をかける
O2 = sigmoid(X2)

# 転置
O2 = np.r_[1, O2]
X3 = np.dot(O2, W2)

print("Anzu Futaba ", ": ", X3)