我创建了以下玩具数据集：
我尝试使用Keras中的神经网络来预测类别：
model = Sequential()model.add(Dense(units=2, activation='sigmoid', input_shape= (nr_feats,)))model.add(Dense(units=nr_classes, activation='softmax'))model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])其中nr_feats和nr_classes设置为2。神经网络只能以50%的准确率预测，返回的结果要么全是1，要么全是2。使用逻辑回归则可以达到100%的准确率。
我无法找出问题出在哪里。
我已经上传了一个笔记本到GitHub，如果你想快速尝试一些东西的话。
编辑1
我大幅增加了训练轮数，准确率从第72轮的0.5开始改善，并在第98轮收敛到1.0。对于这样一个简单的数据库，这仍然显得极其缓慢。
我知道使用单个输出神经元和sigmoid激活函数更好，但我更想了解为什么使用两个输出神经元和softmax激活函数不起作用。
我预处理数据框的方式如下：
from sklearn.preprocessing import LabelEncoderx_train = df_train.iloc[:,0:-1].valuesy_train = df_train.iloc[:, -1]nr_feats = x_train.shape[1]nr_classes = y_train.nunique()label_enc = LabelEncoder()label_enc.fit(y_train)y_train = keras.utils.to_categorical(label_enc.transform(y_train), nr_classes)训练和评估：
model.fit(x_train, y_train, epochs=500, batch_size=32, verbose=True)accuracy_score(model.predict_classes(x_train), df_train.iloc[:, -1].values)编辑2
在将输出层更改为单个神经元和sigmoid激活函数，并使用binary_crossentropy损失函数后，如modesitt所建议的，准确率在200轮内仍保持在0.5，并在100轮后收敛到1.0。
回答：
注意： 如果你想了解真正的原因，请阅读我的回答末尾的“更新”部分。在这种情况下，我提到的另外两个原因只有在学习率设置得非常低（低于1e-3）时才有效。
我整理了一些代码。它与你的代码非常相似，但我只是稍微清理了一下，并让它对我来说更简单。如你所见，我在最后一层使用了一个单元的密集层，并使用了sigmoid激活函数，并且仅将优化器从adam更改为rmsprop（这并不重要，你可以使用adam如果你喜欢）：
import numpy as npimport random# 生成具有两个特征的随机数据n_samples = 200n_feats = 2cls0 = np.random.uniform(low=0.2, high=0.4, size=(n_samples,n_feats))cls1 = np.random.uniform(low=0.5, high=0.7, size=(n_samples,n_feats))x_train = np.concatenate((cls0, cls1))y_train = np.concatenate((np.zeros((n_samples,)), np.ones((n_samples,))))# 打乱数据，因为所有负样本（即类别“0”）都在前面# 然后是所有正样本（即类别“1”）indices = np.arange(x_train.shape[0])np.random.shuffle(indices)x_train = x_train[indices]y_train = y_train[indices]from keras.models import Sequentialfrom keras.layers import Densemodel = Sequential()model.add(Dense(2, activation='sigmoid', input_shape=(n_feats,)))model.add(Dense(1, activation='sigmoid'))model.compile(loss='binary_crossentropy', optimizer='rmsprop', metrics=['accuracy'])model.summary()model.fit(x_train, y_train, epochs=5, batch_size=32, verbose=True)这是输出结果：
Layer (type) Output Shape Param # =================================================================dense_25 (Dense) (None, 2) 6 _________________________________________________________________dense_26 (Dense) (None, 1) 3 =================================================================Total params: 9Trainable params: 9Non-trainable params: 0_________________________________________________________________Epoch 1/5400/400 [==============================] - 0s 966us/step - loss: 0.7013 - acc: 0.5000Epoch 2/5400/400 [==============================] - 0s 143us/step - loss: 0.6998 - acc: 0.5000Epoch 3/5400/400 [==============================] - 0s 137us/step - loss: 0.6986 - acc: 0.5000Epoch 4/5400/400 [==============================] - 0s 149us/step - loss: 0.6975 - acc: 0.5000Epoch 5/5400/400 [==============================] - 0s 132us/step - loss: 0.6966 - acc: 0.5000如你所见，准确率从未从50%增加。如果你将训练轮数增加到50会怎样：
Layer (type) Output Shape Param # =================================================================dense_35 (Dense) (None, 2) 6 _________________________________________________________________dense_36 (Dense) (None, 1) 3 =================================================================Total params: 9Trainable params: 9Non-trainable params: 0_________________________________________________________________Epoch 1/50400/400 [==============================] - 0s 1ms/step - loss: 0.6925 - acc: 0.5000Epoch 2/50400/400 [==============================] - 0s 136us/step - loss: 0.6902 - acc: 0.5000Epoch 3/50400/400 [==============================] - 0s 133us/step - loss: 0.6884 - acc: 0.5000Epoch 4/50400/400 [==============================] - 0s 160us/step - loss: 0.6866 - acc: 0.5000Epoch 5/50400/400 [==============================] - 0s 140us/step - loss: 0.6848 - acc: 0.5000Epoch 6/50400/400 [==============================] - 0s 168us/step - loss: 0.6832 - acc: 0.5000Epoch 7/50400/400 [==============================] - 0s 154us/step - loss: 0.6817 - acc: 0.5000Epoch 8/50400/400 [==============================] - 0s 146us/step - loss: 0.6802 - acc: 0.5000Epoch 9/50400/400 [==============================] - 0s 161us/step - loss: 0.6789 - acc: 0.5000Epoch 10/50400/400 [==============================] - 0s 140us/step - loss: 0.6778 - acc: 0.5000Epoch 11/50400/400 [==============================] - 0s 177us/step - loss: 0.6766 - acc: 0.5000Epoch 12/50400/400 [==============================] - 0s 180us/step - loss: 0.6755 - acc: 0.5000Epoch 13/50400/400 [==============================] - 0s 165us/step - loss: 0.6746 - acc: 0.5000Epoch 14/50400/400 [==============================] - 0s 128us/step - loss: 0.6736 - acc: 0.5000Epoch 15/50400/400 [==============================] - 0s 125us/step - loss: 0.6728 - acc: 0.5000Epoch 16/50400/400 [==============================] - 0s 165us/step - loss: 0.6718 - acc: 0.5000Epoch 17/50400/400 [==============================] - 0s 161us/step - loss: 0.6710 - acc: 0.5000Epoch 18/50400/400 [==============================] - 0s 170us/step - loss: 0.6702 - acc: 0.5000Epoch 19/50400/400 [==============================] - 0s 122us/step - loss: 0.6694 - acc: 0.5000Epoch 20/50400/400 [==============================] - 0s 110us/step - loss: 0.6686 - acc: 0.5000Epoch 21/50400/400 [==============================] - 0s 142us/step - loss: 0.6676 - acc: 0.5000Epoch 22/50400/400 [==============================] - 0s 142us/step - loss: 0.6667 - acc: 0.5000Epoch 23/50400/400 [==============================] - 0s 149us/step - loss: 0.6659 - acc: 0.5000Epoch 24/50400/400 [==============================] - 0s 125us/step - loss: 0.6651 - acc: 0.5000Epoch 25/50400/400 [==============================] - 0s 134us/step - loss: 0.6643 - acc: 0.5000Epoch 26/50400/400 [==============================] - 0s 143us/step - loss: 0.6634 - acc: 0.5000Epoch 27/50400/400 [==============================] - 0s 137us/step - loss: 0.6625 - acc: 0.5000Epoch 28/50400/400 [==============================] - 0s 131us/step - loss: 0.6616 - acc: 0.5025Epoch 29/50400/400 [==============================] - 0s 119us/step - loss: 0.6608 - acc: 0.5100Epoch 30/50400/400 [==============================] - 0s 143us/step - loss: 0.6601 - acc: 0.5025Epoch 31/50400/400 [==============================] - 0s 148us/step - loss: 0.6593 - acc: 0.5350Epoch 32/50400/400 [==============================] - 0s 161us/step - loss: 0.6584 - acc: 0.5325Epoch 33/50400/400 [==============================] - 0s 152us/step - loss: 0.6576 - acc: 0.5700Epoch 34/50400/400 [==============================] - 0s 128us/step - loss: 0.6568 - acc: 0.5850Epoch 35/50400/400 [==============================] - 0s 155us/step - loss: 0.6560 - acc: 0.5975Epoch 36/50400/400 [==============================] - 0s 136us/step - loss: 0.6552 - acc: 0.6425Epoch 37/50400/400 [==============================] - 0s 140us/step - loss: 0.6544 - acc: 0.6150Epoch 38/50400/400 [==============================] - 0s 120us/step - loss: 0.6538 - acc: 0.6375Epoch 39/50400/400 [==============================] - 0s 140us/step - loss: 0.6531 - acc: 0.6725Epoch 40/50400/400 [==============================] - 0s 135us/step - loss: 0.6523 - acc: 0.6750Epoch 41/50400/400 [==============================] - 0s 136us/step - loss: 0.6515 - acc: 0.7300Epoch 42/50400/400 [==============================] - 0s 126us/step - loss: 0.6505 - acc: 0.7450Epoch 43/50400/400 [==============================] - 0s 141us/step - loss: 0.6496 - acc: 0.7425Epoch 44/50400/400 [==============================] - 0s 162us/step - loss: 0.6489 - acc: 0.7675Epoch 45/50400/400 [==============================] - 0s 161us/step - loss: 0.6480 - acc: 0.7775Epoch 46/50400/400 [==============================] - 0s 126us/step - loss: 0.6473 - acc: 0.7575Epoch 47/50400/400 [==============================] - 0s 124us/step - loss: 0.6464 - acc: 0.7625Epoch 48/50400/400 [==============================] - 0s 130us/step - loss: 0.6455 - acc: 0.7950Epoch 49/50400/400 [==============================] - 0s 191us/step - loss: 0.6445 - acc: 0.8100Epoch 50/50400/400 [==============================] - 0s 163us/step - loss: 0.6435 - acc: 0.8625