QA Official

java's Basic Data Structure

https://qaofficial.com/post/2019/04/07/69344-javaamp#x27s-basic-data-structure.html 2019-04-07
java Data Structure Introduction Data structure is the way that computers store and organize data.A data structure refers to a collection of data elements that have one or more specific relationships with each other.In general, carefully selected data structures can lead to higher operation or storage efficiency.Data structures are often related to efficient retrieval algorithms and indexing techniques. The data structures commonly used in Java are basically Array array, Collection and Map Collection.

keras installation, win7+ubuntu dual system, gpu, fast.ai, theano

https://qaofficial.com/post/2019/04/07/69417-keras-installation-win7-ubuntu-dual-system-gpu-fast.ai-theano.html 2019-04-07
originally wanted to use fast.ai as an introduction.It is said that it was very fast, but after studying for several days, the program could not understand it. So going around made a big circle and finally returned to keras.Learn keras first and then fast.ai Note here the steps for installing keras (20170625) 1 ubuntu installed Programmers who really enter deep learning should install one.This is not a virtual machine. It

Actual Combat Learning in Tensor Flow (XVII) [Natural Language Processing, RNN, LSTM]

https://qaofficial.com/post/2019/04/07/69419-actual-combat-learning-in-tensor-flow-xvii-natural-language-processing-rnn-lstm.html 2019-04-07
Natural Language Processing (NLP) problems are serialized.Feedforward neural network processes incoming data in a single feed forward, assuming that all inputs are independent and the mode is lost.Recurrent neural network,RNN) explicitly models the neural network for time.RNN neurons can receive weighted inputs from other neurons.RNN neurons can establish connections with higher layers or lower layers.Implicit activity values are memorized between adjacent inputs of the same sequence.LSTM 2006.Speech recognition, speech synthesis, handwritten conjoined character recognition, time series prediction, image title generation, end-to-end machine translation.

Add a New loss Definition to CAFFE (Convolutional Architecture for Fast Feature Embedding)

https://qaofficial.com/post/2019/04/07/69310-add-a-new-loss-definition-to-caffe-convolutional-architecture-for-fast-feature-embedding.html 2019-04-07
* Key: The input and output of forward/backward is What?* Define loss Function take CCC-classification problem as an example, define softmax_l2_loss, input sample XXX, and the loss of k∈[0,C]k∈[0,C]k \in [0,C] output is defined as L_k(X) = (y_k - f_k(X))^2 Where ykyky_k is the KKKth element of label YYY, fk(X)fk(X)f_k(X) is the input xixix_i, and the output KKKth value is defined as softmax function f_k(X) = \frac{e^{z_k}}{\sum_j{e^{z_j}}} Zkzkz_k is the original output value of the network.

Common Class Comparison of Set Class Framework

https://qaofficial.com/post/2019/04/07/69372-common-class-comparison-of-set-class-framework.html 2019-04-07
When designing software with a " set framework", it is useful to remember the following hierarchical relationships of the framework's four basic interfaces: The Collection interface is a set of objects that allow duplication. Set interface inherits Collection, but duplication is not allowed. List interface inherits Collection, allows duplication, and introduces position subscript. Map inherits neither Set nor Collection and accesses key-value pairsLet's use the following chart to describe the

Comparison of Common Collection Classes (Set, Map, List)

https://qaofficial.com/post/2019/04/07/69345-comparison-of-common-collection-classes-set-map-list.html 2019-04-07
The two structures at the bottom of the data storage mode in Java: array and linked list. The characteristics of array are continuous space and fast addressing. However, when deleting or adding elements, there is a large movement, so the query speed is fast and the addition and deletion are slow.On the other hand, linked lists are just the opposite. Due to discontinuous space and difficult addressing, adding and deleting

Comparison of Several Pairs of Set Classes

https://qaofficial.com/post/2019/04/07/69368-comparison-of-several-pairs-of-set-classes.html 2019-04-07
JDK has already provided us with collection classes when developing Java.The following are respectively introduced: Collection ├List │├LinkedList │├ArrayList │└Vector │ └Stack └Set Map ├Hashtable ├HashMap └WeakHashMap Collection interface Collection is the most basic Collection and represents a group of Elements.It is necessary to know each class: whether the same elements are allowed and whether they can be sorted.The Java SDK does not provide classes that directly inherit from the Collection.

Deep Understanding of JVM (1) JVM Memory Model

https://qaofficial.com/post/2019/04/07/69421-deep-understanding-of-jvm-1-jvm-memory-model.html 2019-04-07
The Java Virtual Machine will divide the memory it manages into several different data areas during the execution of Java programs, including the following runtime data areas in total. 1, Program Counter Register program counter is a small memory space, its function is: 1.1. Can be seen as a signal indicator of bytecode executed by the current thread.Byte code interpreter selects the next bytecode instruction to be executed by changing the value of the counter.

Getting Started with Tensor Flow: mnist

https://qaofficial.com/post/2019/04/07/69387-getting-started-with-tensor-flow-mnist.html 2019-04-07
1. acquisition and use of mnist data set1.1 get mnistfrom tensorflow.contrib.learn.python.learn.datasets.mnist import read_data_sets mnist = read_data_sets("MNIST_data/", one_hot=True) # "MNIST_data/"是数据保存的位置,one-hot是否把label变成one-hot编码 1.2 use mnistmnist.train.next_batch(50) #

Introduction of Dropout Principle in++Deep Learning

https://qaofficial.com/post/2019/04/07/69430-introduction-of-dropout-principle-in-deep-learning.html 2019-04-07
If the problem with the network is that it is densely connected, it will be forced to be sparse for so long. The algorithm based on this idea is dropout algorithm. 1: Introduction Because in some models of machine learning, if the parameters of the model are too many and the training samples are too few, then the trained model is prone to over-fitting.One of the problems often encountered in