ニューラルネットワークの分類問題

Question

ゼロから作るDeep Learning（オライリー・ジャパン）のchapter4 Mnistのクラス分類のニューラルネットワークを改良して、scikit-learn のワインデータを分類しようとしましたが、学習がすすみません。モデルや、逆伝播のロジックは本のままで、投入データの部分を改良しています。どこが間違っているのでしょうか。
学習後の重みパラメータ"W2"は、ありえない値になっしまいます。

import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn import linear_model, datasets

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

def sigmoid_grad(x):
    return (1.0 - sigmoid(x)) * sigmoid(x)

def softmax(x):
    if x.ndim == 2:
        x = x.T
        x = x - np.max(x, axis=0)
        y = np.exp(x) / np.sum(np.exp(x), axis=0)
        return y.T 

    x = x - np.max(x) 
    return np.exp(x) / np.sum(np.exp(x))

def cross_entropy_error(y, t):
    if y.ndim == 1:
        t = t.reshape(1, t.size)
        y = y.reshape(1, y.size)
        
    # 教師データがone-hot-vectorの場合、正解ラベルのインデックスに変換
    if t.size == y.size:
        t = t.argmax(axis=1)
             
    batch_size = y.shape[0]
    return -np.sum(np.log(y[np.arange(batch_size), t] + 1e-7)) / batch_size

class TwoLayerNet:

    def __init__(self, input_size, hidden_size, output_size, weight_init_std=0.01):
        # 重みの初期化
        self.params = {}
        self.params['W1'] = weight_init_std * np.random.randn(input_size, hidden_size)
        self.params['b1'] = np.zeros(hidden_size)
        self.params['W2'] = weight_init_std * np.random.randn(hidden_size, output_size)
        self.params['b2'] = np.zeros(output_size)

    def predict(self, x):
        W1, W2 = self.params['W1'], self.params['W2']
        b1, b2 = self.params['b1'], self.params['b2']
    
        a1 = np.dot(x, W1) + b1
        z1 = sigmoid(a1)
        a2 = np.dot(z1, W2) + b2
        y = softmax(a2)
        
        return y
        
    # x:入力データ, t:教師データ
    def loss(self, x, t):
        y = self.predict(x)
        
        return cross_entropy_error(y, t)
    
    def accuracy(self, x, t):
        y = self.predict(x)
        y = np.argmax(y, axis=1)
        t = np.argmax(t, axis=1)
        
        accuracy = np.sum(y == t) / float(x.shape[0])
        return accuracy
        
    # x:入力データ, t:教師データ
    def gradient(self, x, t):
        W1, W2 = self.params['W1'], self.params['W2']
        b1, b2 = self.params['b1'], self.params['b2']
        grads = {}
        
        batch_num = x.shape[0]
        
        # forward
        a1 = np.dot(x, W1) + b1; #print("a1",end=":");print(a1)
        z1 = sigmoid(a1); #print("z1",end=":");#print(z1)
        a2 = np.dot(z1, W2) + b2;#print("a2",end=":");print(a2)
        y = softmax(a2);#print("y",end=":");print(y)
        
        # backward
        dy = (y - t) / batch_num; 
        grads['W2'] = np.dot(z1.T, dy); #print("np.dot(z1.T, dy): " + str(grads['W2']))
        grads['b2'] = np.sum(dy, axis=0); #print("np.sum(dy, axis=0)" + str(grads['b2']))
        
        da1 = np.dot(dy, W2.T); #print("np.dot(dy, W2.T)" + str(da1))
        dz1 = sigmoid_grad(a1) * da1 ; #print("sigmoid_grad(a1) * da1" + str(dz1))
        grads['W1'] = np.dot(x.T, dz1) ; #print("np.dot(x.T, dz1)" + str(grads['W1']))
        grads['b1'] = np.sum(dz1, axis=0) ; #print("np.sum(dz1, axis=0)" + str(grads['b1']))
        return grads

# データの読み込み
load_data = datasets.load_wine()
features = pd.DataFrame(data=load_data.data, columns=load_data.feature_names)
targets = pd.DataFrame(data=load_data.target, columns=['class'])
x_train, x_test, t_train, t_test = train_test_split(features, targets, test_size=0.1)

network = TwoLayerNet(input_size=13, hidden_size=100, output_size=4)

iters_num = 1000  # 繰り返しの回数を適宜設定する
train_size = x_train.shape[0]
batch_size = 8
learning_rate = 0.01

train_loss_list = []
train_acc_list = []
test_acc_list = []

iter_per_epoch = max(train_size / batch_size, 1)

cnt = 0
for i in range(iters_num):
    cnt+=1
    batch_mask = np.random.choice(train_size, batch_size)
    x_batch = x_train.values[batch_mask]
    t_batch = t_train.values[batch_mask]

    # 勾配の計算
    grad = network.gradient(x_batch, t_batch)
    
    # パラメータの更新
    for key in ('W1', 'b1', 'W2', 'b2'):
        network.params[key] -= learning_rate * grad[key]
    
    loss = network.loss(x_batch, t_batch);#print("loss: " + str(loss))
    train_loss_list.append(loss)
    print(loss)

    if i % iter_per_epoch == 0 and True:
        train_acc = network.accuracy(x_train.values, t_train.values)
        test_acc = network.accuracy(x_test.values, t_test.values)
        train_acc_list.append(train_acc)
        test_acc_list.append(test_acc)
        print("train acc, test acc | " + str(train_acc) + ", " + str(test_acc))
print("W1")
print(network.params["W1"])
print("b1")
print(network.params["b1"])
print("W2")
print(network.params["W2"])
print("b2")
print(network.params["b2"])

# グラフの描画
markers = {'train': 'o', 'test': 's'}
x = np.arange(len(train_acc_list))
plt.plot(x, train_acc_list, label='train acc')
plt.plot(x, test_acc_list, label='test acc', linestyle='--')
plt.xlabel("epochs")
plt.ylabel("accuracy")
plt.ylim(0, 1.0)
plt.legend(loc='lower right')
plt.show()

Answer 1

データ読み込み後、正規化と、One-hot-vector処理をすると、うまくいきました。

for each in features.columns:
    mean, std = features[each].mean(), features[each].std()
    features.loc[:, each] = (features[each] - mean)/std

targets = pd.get_dummies(targets['class'], prefix='class')