我在Python中使用sklearn的MLPClassifier构建了一个用于分类任务的神经网络。我希望绘制一个准确率与轮次（epochs）的曲线，以了解需要多少轮次才能达到一定的准确率。我能找到的唯一方法是使用partial_fit()在一个循环中。这里是实现这个功能的代码:
from sklearn.preprocessing import StandardScalerfrom sklearn.decomposition import PCAfrom sklearn.neural_network import MLPClassifierimport pandas as pdimport numpy as npimport matplotlib.pyplot as pltscaler = StandardScaler()scaler.fit(df_train_sample)X_train = scaler.transform(df_train_sample)scaler.fit(df_val)X_val = scaler.transform(df_val)pca = PCA(pca_frac)pca.fit(X_train)X_train = pca.transform(X_train)X_val = pca.transform(X_val)n_classes = np.unique(labels_train_sample)n_train_sample = len(df_train_sample)scores_train = []scores_val = []epoch = 0while epoch < max_iter: random_perm = np.random.permutation(n_train_sample) mini_batch_index = 0 while True: indices = random_perm[mini_batch_index:mini_batch_index + batch_size] mlpc.partial_fit(X_train[indices], labels_train_sample[indices], classes=n_classes) mini_batch_index += batch_size if mini_batch_index >= n_train_sample: break scores_train.append(mlpc.score(X_train, labels_train_sample)) scores_val.append(mlpc.score(X_val, labels_val)) epoch += 1fig, ax = plt.subplots()ax.plot(np.arange(1, max_iter + 1), scores_train, label = "Train")ax.plot(np.arange(1, max_iter + 1), scores_val, label = "Validation")这里，max_iter是轮次数，mlpc是分类器，定义如下:
seed = 123hidden_layers = [30, 15]activation = "relu"learning_rate = 5e-4beta_1 = 0.99epsilon = 1e-4batch_size = 200 max_iter = 200 tol = 1e-4warm_start = Trueshuffle = Truemlpc = MLPClassifier( hidden_layer_sizes = hidden_layers, activation = activation, batch_size = batch_size, learning_rate_init = learning_rate, beta_1 = beta_1, epsilon = epsilon, warm_start = warm_start, shuffle = shuffle, max_iter = max_iter, tol = tol, random_state = seed)为了确保万无一失，这里是如何从原始数据框构造df_train_sample和labels_train_sample的:
df_train_sample = df_train.sample(N, replace = False).reset_index(drop = True)labels_train_sample = labels_train[df_train_sample.index].reset_index(drop = True)其中N是采样的行数。df_val和labels_val是验证数据，直接从.csv文件中读取，没有做任何修改。请注意，标签是布尔值。
问题是，如果使用mlpc.fit()，算法在采样数据集上的准确率约为82%，而我发布的代码片段的准确率为65%。这是图表:
在网上搜索后，我发现打乱数据可能会有所帮助，但如您所见，每个轮次数据都已经打乱了。这是为什么呢？还有没有其他更直接的方法来构建上述图表呢？
回答：
我已经找到了问题所在。问题不在于partial_fit()，而是我构建样本数据框的方式:
df_train_sample = df_train.sample(N, replace = False).reset_index(drop = True)labels_train_sample = labels_train[df_train_sample.index].reset_index(drop = True)在这一部分，我在构建df_train_sample时重置了其索引，但随后我使用其索引从labels_train中采样相应的行。如果我不重置索引（这是我之前版本中使用的做法），这将有效。
解决方案只是在重置索引之前存储索引，像这样:
df_train_sample = df_train.sample(N, replace = False)train_index = df_train_sample.indexdf_train_sample = df_train_sample.reset_index(drop = True)labels_train_sample = labels_train[train_index].reset_index(drop = True)