Scikit-learn MLPRegressor - 如何不预测负面结果？

Posted 2023-03-12

【中文标题】Scikit-learn MLPRegressor - 如何不预测负面结果？

【英文标题】：Scikit-learn MLPRegressor - How not to predict negative results?

【发布时间】：2018-01-31 00:06:50

【问题描述】：

我正在尝试使用 MLPRegressor 训练和测试我的数据集。我有两个数据集（训练数据集和测试数据集），它们都具有完全相同的特征和标签列。这是我的数据集示例：

Full,Id,Id & PPDB,Id & Words Sequence,Id & Synonyms,Id & Hypernyms,Id & Hyponyms,Gold Standard
1.667,0.476,0.952,0.476,1.429,0.952,0.476,2.345
3.056,1.111,1.667,1.111,3.056,1.389,1.111,1.9
1.765,1.176,1.176,1.176,1.765,1.176,1.176,2.2
0.714,0.714,0.714,0.714,0.714,0.714,0.714,0.0
................

这是我的代码：

import pandas as pd
import numpy as np 

from sklearn.neural_network import MLPRegressor

randomseed = np.random.seed(0)

datatraining = pd.read_csv("datatrain.csv")

datatesting = pd.read_csv("datatest.csv")

columns = ["Full","Id","Id & PPDB","Id & Words Sequence","Id & Synonyms","Id & Hypernyms","Id & Hyponyms"]

labeltrain = datatraining["Gold Standard"].values
featurestrain = datatraining[list(columns)].values


labeltest = datatesting["Gold Standard"].values
featurestest = datatesting[list(columns)].values

X_train = featurestrain
y_train = labeltrain

X_test = featurestest
y_test = labeltest

mlp = MLPRegressor(solver='lbfgs', hidden_layer_sizes=50, max_iter=1000, learning_rate='constant', random_state=randomseed)

mlp.fit(X_train, y_train)

print('Accuracy training : :.3f'.format(mlp.score(X_train, y_train)))
print

predicting = mlp.predict(X_test)
print predicting
print

这是预测的结果：

[ 1.97553444  3.43401776  3.04097607  2.7015464   2.03777686  3.63274593
  3.37826962 -0.60260337  0.41626517  3.5374289   3.66114929  3.244683
  2.6313756   2.14243075  3.20841434  2.105238    4.9805092   4.00868273
  2.45508505  4.53332828  3.41862096  3.35721078  3.23069344  3.72149434
  4.9805092   2.61705563  1.55052494 -0.14135979  2.65875196  3.05328206
  3.51127424  0.51076396  2.39947967  1.95916595  3.71520651  2.1526807
  2.26438616  0.73249057  2.46888695  3.56976227  1.03109988  2.15894353
  2.06396103  0.66133707  4.72861602  2.4592647   2.84176811  2.3157664
  1.68426416  2.56022955 -0.00518545  1.67213609  0.6998739   3.25940136
  3.25369266  3.88888542  1.9168694   2.26036302  3.97917769  2.00322903
  3.03121106  3.29083723  0.6998739   4.33375678  0.6998739   2.71141538
 -4.23755447  3.958574    2.67765274  2.68715423  2.32714117  2.6500056
  ........]

我们可以看到，有一些负面的结果。如何不预测负面结果？此外，我的数据集包含所有正值。

【问题讨论】：

您需要以一种或另一种方式对预测值施加积极性限制。因此，您可能希望将您的问题从 为什么显示负面结果 转变为 如何不预测负面结果，或者更一般地说，转变为 如何限制域预测.

【参考方案1】：

假设您没有分类变量。此外，您在问题中提到您拥有所有积极的价值观。 尝试使用SatandardSacler() 标准化您的数据。使用您的 X_train 和 y_train 到 standardize 数据。

from sklearn import preprocessing as pre
...
scaler = pre.StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.fit_transform(X_test)

根据您的情况使用最佳参数初始化模型后，fit 缩放数据，

mlp.fit(X_train_scaled, y_train)
...
predicting = mlp.predict(X_test_scaled)

应该这样做。让我知道事情的后续。

还有一些不错的读物，

https://stats.stackexchange.com/questions/189652/is-it-a-good-practice-to-always-scale-normalize-data-for-machine-learning https://stats.stackexchange.com/questions/7757/data-normalization-and-standardization-in-neural-networks

【参考方案2】：

添加带有一个 ReLU 节点的第二个隐藏层。

【讨论】：

欢迎来到 ***。考虑阅读此说明How to answer