The full text 1716 words, when learning is expected to grow 5 Fenzhong
Source: ijava
There is evidence that the neural network machine learning algorithm is superior to many areas. The latter continue to learn, until the best set of features that satisfy the predicted value.
However, the neural network variable zoom into a series of numbers, once it completes the learning phase, for us, it becomes impossible to distinguish the features.
If you only consider the fact of prediction, neural network algorithm that has been used. But the industry needs to model the environment, it can be given for stakeholders characteristic or variable significance. These stakeholders can be anyone, not just learning or machine learning to understand the depth of human knowledge.
What is the biggest difference between Random Forests and Neural networks are?
Random Forests and Neural networks are different technologies, different learning styles, but can be used in a similar field. Random Forests are machine learning technology, and neural network technology is the depth of learning.
What is a neural network?
A neural network is a network computing model, which substantially mimics the human brain function, and can replicate human thinking and perception of the same way. The neural network is a hierarchy of interconnected nodes, which nodes comprise network activation function calculating output.
Neural networks are another method of machine learning, computer training by analyzing the samples to learn to perform tasks. Because the neural network is substantially simulate the human brain, it will include thousands of nodes connected to each other. A node can be connected to several nodes which received lower layer data and upper layer several nodes which receive data. Each input data point receiving a weight, it is subject to arithmetic operations. If the weighting is equal to zero, a deviation will increase, and then passed to the activation function.
Neural Networks
There are three basic neural network structure
1. Before a single layer feedforward network
· It is an extension of Perceptron, is the most simple network. Additional hidden node between the input and output layers.
2. Before the multi-layer feedforward network
· In addition to the input and output, this type of network there is one or more hidden layers. It is responsible for data transfer between the input and output layers.
3. Recurrent Network
· Recurrent neural network with both similar to the above, but is widely used in predictive text and time series data. The most famous recurrent neural network is the "memory length" model (LSTM).
What is a Random Forests?
随机森林是决策树的集合，因此，最终节点/叶节点将是分类问题的多数类或回归问题的平均类。
一个随机森林会长出许多分类树，对于树的每个输出，称之为该树为该类的“投票”。树是这么生长起来的：
1. 每棵树行的随机样本从训练数据中抽取。
2. 从步骤1选出样本后，利用特征子集在选出的树上进行拆分。
3. 每棵树都按照参数指定的最大范围生长，直到它对类进行投票。
为什么推荐随机森林？
使用随机森林而非决策树的根本，是将多个决策树的预测组合成一个模型。逻辑就是一个由许多平庸的模型组成的模型，仍然优于一个单一的好模型。看看随机森林的主流表现，就知道这是有道理的。因此，随机森林不易产生过拟合。
在决策树这样的柔性模型中，可能会出现过拟合，模型会记住训练数据，并学习数据中的任何干扰。这将使它无法预测测试数据。
随机森林可以将多棵树组合成一个集合模型，这能降低决策树等柔性模型的高方差。
来源：tuxi
相比于神经网络，什么时候该使用随机森林？
随机森林计算成本较低，不需要依靠GPU完成培训。随机森林可以提供决策树的不同解释，并且具有更好的性能。神经网络需要普通人手头更多的数据才能真正有效。神经网络只会简单地破坏特征的可解释性，以至于为了性能而变得毫无意义。虽然听起来有些道理，但还是要看每个项目具体分析。
如果目标是创建预测模型，不考虑变量的影响，强烈推荐神经网络，但是需要资源来实现这一点。如果需要了解变量，那么不管个人喜好，在这种情况下通常会发生的情况是：性能受到轻微的影响，以确保仍然可以了解每个变量是如何为预测模型作贡献。
如果你还有什么补充，欢迎在评论区留言哟~
留言 点赞 关注
我们一起分享AI学习与发展的干货
欢迎关注全平台AI垂类自媒体 “读芯术”
（添加小编微信：dxsxbb，加入读者圈，一起讨论最新鲜的人工智能科技哦～）
