QA Official

[ keras ] summary of cat-dog war

https://qaofficial.com/post/2019/04/29/23926-keras-summary-of-cat-dog-war.html 2019-04-29
This blog mainly refers to " Building Image Classification Model for Small Datasets" in keras official documents.This article records the problems encountered in studying this article and some methods explored by myself.The general idea of the official document is: Firstly, the VGG-16 network with the full connection layer removed is used to obtain the bottleneck of the data set.Feature, then design several full connection layers, train them, and finally fine-tune the last few convolution layers.

kaggle Project Actual Combat-Cat and Dog Classification Detection

https://qaofficial.com/post/2019/04/29/23937-kaggle-project-actual-combat-cat-and-dog-classification-detection.html 2019-04-29
Main Reference: Deep Learning: Detailed Explanation and Actual Combat of ——caffe's Classic Model kaggle dataset download link: https://www.kaggle.com/c/dogs-vs-cats-redux-kernels-edition/data Data Set Description: consists of two files: train.zip: training set, picture file naming format: cat.X.jpg,dog.X.jpg this is a two-classification problem, which requires 0/1 classification according to cat/dog category, i.e. cat->0,dog->1 test.zip: used for detection, naming format: X.jpg, used to verify the recognition accuracy of

keras Migration Learning, Fine-tuning, model's predict Function Definition

https://qaofficial.com/post/2019/04/29/23933-keras-migration-learning-fine-tuning-model#39s-predict-function-definition.html 2019-04-29
click here: cat and dog vs keras instance def add_new_last_layer(base_model, nb_classes): """Add last layer to the convnet Args: base_model: keras model excluding top nb_classes: # of classes Returns: new keras model with last layer """ x = base_model.output x = GlobalAveragePooling2D()(x) x = Dense(FC_SIZE, activation='relu')(x) predictions = Dense(nb_classes, activation='softmax')(x) model = Model(input=base_model.input, output=predictions) return model Load the pre-training model as the front-end network and fine-tune it on its own data

keras-transfer learning fine-tuning

https://qaofficial.com/post/2019/04/29/23932-keras-transfer-learning-fine-tuning.html 2019-04-29
This program demonstrates the process of fine-tuning a pre-trained model on a new data set.We freeze the convolution layer and only adjust the full connection layer. train a convolution network on the MNIST dataset using the first five digits [ 0 ... 4 ]. In the last five digits [ 5 ... 9 ], the convolution network is used to classify, freeze the convolution layer and fine-tune the full connection

using resnet to do kaggle cat and dog war image recognition, 98 accuracy in seconds

https://qaofficial.com/post/2019/04/29/23912-using-resnet-to-do-kaggle-cat-and-dog-war-image-recognition-98-accuracy-in-seconds.html 2019-04-29
1, Data Introduction this data set comes from Kaggle, with 12,500 cats and 12,500 dogs.Here is a brief introduction to the overall idea1.1 Train a small network directly from the picture (as a reference method), that is, the ordinary cnn methodAfter 2, 2, I will use the latest pre-trained resnet and other methods for training 2 Data Promotion and cnn In order to make the best use of our limited

word2vec-(1) nltk implements simple word cutting, sentiment analysis, and text similarity (TF-IDF)

https://qaofficial.com/post/2019/04/29/24014-word2vec-1-nltk-implements-simple-word-cutting-sentiment-analysis-and-text-similarity-tf-idf.html 2019-04-29
Nltk from nltk.corpusimport brown (1) Brown. Categories () The article directory under this file (2) len(brown.sents()) (3) len(brown.words()) tokenizer participle nltk.tokenize(sentence) stuttering participleThree Word Cutting Modes Import jieba jieba.cut (' open the official one', cut _ all = true) # full mode jieba.cut (' the official one', cut _ all = false) # exact mode print "Full Mode:", "/".join(seg_list) seg _ list = jieba.cut _ for _ search (" Xiao

CGLIB Introduction and Principle

https://qaofficial.com/post/2019/04/28/58796-cglib-introduction-and-principle.html 2019-04-28
CGLIB Introduction and Principle (Some Excerpts from Network) 1. What is CGLIB? CGLIB is a powerful and high-performance code generation package.It provides proxies for classes that do not implement interfaces and a good supplement for JDK's dynamic proxies.You can usually use Java's dynamic proxy to create a proxy, but CGLIB is a good choice when the class you want to proxy does not implement an interface or for better performance.

Hbase getting started

https://qaofficial.com/post/2019/04/28/68958-hbase-getting-started.html 2019-04-28
There are many big data frameworks. I learned some hive before. Because it has some shortcomings such as slow response and does not support transactions, I still need to know about hbase. First look at what others have summarized. HBase detailed overview HBase is deep and shallow.

Java Collection Class: Difference and Application of Set, List, Map and Queue

https://qaofficial.com/post/2019/04/28/69356-java-collection-class-difference-and-application-of-set-list-map-and-queue.html 2019-04-28
Java Collection Class Basic Concepts In programming, it is often necessary to store multiple data centrally.Traditionally, arrays are a good choice for us, provided that we already know the number of objects we will save in advance.Once the array length is specified at the initialization of the array, the array length is immutable. If we need to save a data that can grow dynamically (the specific amount cannot be determined

Java String class trap analysis

https://qaofficial.com/post/2019/04/28/69650-java-string-class-trap-analysis.html 2019-04-28
1. For the equals () method of String class, it determines whether the current string is consistent with the contents of the incoming string. 2. For the equality judgment of String objects, please use the equals () method.Instead of using = =. 3 and String are constants whose objects cannot be changed once they are created.When using+concatenation Strings, a new String object is generated instead of appending content to the