Question

我有一个基础神经网络，我在Keras训练过。我正在研究学习速度和动量项的影响，我想绘制一个漂亮的三维图形，以显示学习速度和动量对准确性的影响。

我设法使用示例代码成功绘制了一个trisurf图，但每当我使用自己的数据时，我都会遇到错误。这些示例似乎使用了大约1000个值的numpy数组，而我只有大约6个不同的学习速率和动量值，给出了大小为6,6和36的numpy数组。当我尝试使用这些值绘制图形时，我得到了以下错误：

RuntimeError：qhull Delaunay三角测量计算错误：单数输入数据（exitcode = 2）

我不理解这个错误消息，以及为什么它对示例数据起作用，而不是我自己的。有什么建议吗？

我的代码如下：

momentum_terms = np.array([0.00001,0.0001,0.001,0.01, 0.1, 1])
learning_rates = np.array([0.00001,0.0001,0.001,0.01, 0.1, 1])
train_accuracies = np.empty([36])
test_accuracies = np.empty([36])
for learning_rate in learning_rates:
    for momentum in momentum_terms:
        model = Sequential()
        model.add(Dense(18, activation='relu', input_shape = (2,)))
        model.add(Dense(18, activation='relu'))
        model.add(Dense(1, activation='sigmoid'))
        model.summary()

        model.compile(loss='binary_crossentropy',
                      optimizer=SGD(lr = learning_rate, momentum = momentum),
                      metrics=[binary_accuracy])

        history = model.fit(x_train, y_train,
                            batch_size=batch_size,
                            epochs=epochs,
                            verbose=1,
                            validation_data=(x_test, y_test))
        score = model.evaluate(x_test, y_test, verbose=0)
        np.append(train_accuracies, history.history['binary_accuracy'][-1] * 100)
        np.append(test_accuracies, history.history['val_binary_accuracy'][-1] * 100)

x = momentum_terms
y = learning_rates
z = test_accuracies

ax = plt.axes(projection='3d')
ax.plot_trisurf(x, y, z, cmap='viridis', edgecolor='none');
plt.show()

Answer 1

您没有提供足够的数据来生成3d图（see this related SO question）。不需要传递6,6和36，而是需要传递36,36和36.重做代码，以便在准确的情况下将每对动量项和学习速率项存储在循环中。

所以你应该：

x = [0.00001,0.00001,0.00001,0.00001,0.00001， 0.00001,0.0001， 0.0001，....]从学习率选择中获得36个值

y = [0.00001,0.0001,0.001,0.01,0.1， 1,0.00001， 0.0001,0.001,0.01,0.1,1 ......]，36个值来自动量选择< / p>

z =上述每个组合的36个精度数组

学习率，动量和准确性的3d图

1 个答案: