这是我正在处理的代码:
from __future__ import print_functionfrom keras.models import Sequentialfrom keras.layers import Densefrom sklearn.cross_validation import train_test_splitimport numpynumpy.random.seed(7)data_pixels=np.genfromtxt("pixels_dataset.csv", delimiter=',')classes_dataset=np.genfromtxt("labels.csv",dtype=np.str , delimiter='\t')x_train, x_test, y_train, y_test = train_test_split(data_pixels, classes_dataset, test_size=0.3x_train 的形状为 (1708, 3072)
array([[ 0., 0., 0., ..., 0., 0., 0.], [ 0., 0., 0., ..., 1., 1., 1.], [ 1., 1., 1., ..., 1., 1., 1.], ..., [ 0., 0., 0., ..., 1., 1., 1.], [ 1., 1., 1., ..., 1., 1., 1.], [ 0., 0., 0., ..., 1., 1., 1.]])y_train 的形状为 (1708,)
array(['7', 'f', '3', ..., '6', 'o', 'O'], dtype='|S5')y_train 中的字符包括
: , : ; ! è à ä Aa..Zz 0-9
model = Sequential()model.add(Dense(12, input_dim=3072, activation='relu'))model.add(Dense(8, activation='relu'))model.add(Dense(1, activation='sigmoid'))model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])执行以下代码后我得到了错误:
model.fit(x_train,y_train, epochs=150, batch_size=10)错误信息为
ValueError: could not convert string to float: A我尝试了以下替代方案 :1)
x_train=n.array(x_train)y_train=n.array(y_train)2)
model.fit(x_train,str(y_train), epochs=150, batch_size=10)但我得到了同样的错误。然后我尝试了另一个替代方案
from sklearn.preprocessing import LabelBinarizerencoder = LabelBinarizer()y_train = encoder.fit_transform(y_train)然后我得到了一个新的错误:
ValueError: Error when checking model target: expected dense_21 to have shape (None, 1) but got array with shape (1708, 66)回答：
修改以下代码行:
model.add(Dense(66, activation='softmax'))和:
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])问题在于你想预测一个编码为长度为66的one-hot向量的char。在这种情况下，你需要设置输出层为所需的长度，并使用categorical_crossentropy损失函数和softmax激活函数。
