Personal developer portfolio and technical blog by Gary.https://justinqy.github.io/https://justinqy.github.io/blog/interview-self-introduction-bq/https://justinqy.github.io/blog/interview-self-introduction-bq/Interview preparation Self-Introduction Before graduates and during graduates. Behavior Situations Won a bronze medal in a programming competition with teammates Fit questions: teaWed, 29 Oct 2025 16:29:04 GMTInterviewhttps://justinqy.github.io/blog/ai-rag-rag-from-scratch/https://justinqy.github.io/blog/ai-rag-rag-from-scratch/LangChain Retrieval Augmentation Generation Build RAG from Scratch There are several steps to build a RAG system from scratch. 1. Loading documents. 2. Chunk the documents into splWed, 24 Sep 2025 15:15:01 GMThttps://justinqy.github.io/blog/ai-interview-ibminterviewrecords/https://justinqy.github.io/blog/ai-interview-ibminterviewrecords/Interview Questions by IBM Data Scientist Intern Self Introduction Part Introduce yourself Your favorite programming languages, why? C++, Python. Data Science and Machine LearningThu, 28 Aug 2025 16:05:39 GMTInterviewhttps://justinqy.github.io/blog/ai-interview-coop-interview-prepare/https://justinqy.github.io/blog/ai-interview-coop-interview-prepare/Interview preparation for 2025 September co-op positions. Government of Ontario - Data Engineer (Co-op) SQL ETL 1. What's ETL? ETL is Extract , Transform and Load . It’s the most iTue, 01 Jul 2025 16:30:16 GMTInterviewhttps://justinqy.github.io/blog/ai-projects-langchain/https://justinqy.github.io/blog/ai-projects-langchain/LangChain产品线以及文档拆分, 向量化技术栈总结 Use OpenAI's Models for this document. LangChain LangChain Initial a chat model Embedding Vector Store LangGraph Agent = control flow defined by an LLMMon, 16 Jun 2025 10:36:00 GMThttps://justinqy.github.io/blog/ai-projects-rag-qabot/https://justinqy.github.io/blog/ai-projects-rag-qabot/RAG based Question-Answering System Loading Documents DocumentLoaders WebBaseLoaderWed, 11 Jun 2025 12:22:16 GMTRAGhttps://justinqy.github.io/blog/cloud-aws-services-aws-basic-services/https://justinqy.github.io/blog/cloud-aws-services-aws-basic-services/Basic Services in AWS IAM (Identity and Access Management) 1. users & groups management 2. Access to AWS AWS Management Console (Through Browser, username + password + MFA) AWS ComMon, 02 Jun 2025 16:08:04 GMTAWShttps://justinqy.github.io/blog/ai-machine-learning-recommender-system/https://justinqy.github.io/blog/ai-machine-learning-recommender-system/Algorithms for Recommender Systems Collaborative Filtering Algorithm (协同过滤) --- Building Recommender System with per-item features For example, each movie has 2 features: $x 1$ isWed, 28 May 2025 16:28:48 GMThttps://justinqy.github.io/blog/ai-machine-learning-algo-for-unsupervised/https://justinqy.github.io/blog/ai-machine-learning-algo-for-unsupervised/Algorithms for unsupervised learning Clustering Practical application in clustering: 1. Grouping similar news 2. Market segmentation 3. DNA analysis K-means Algorithm 'The K-meansMon, 26 May 2025 15:43:18 GMThttps://justinqy.github.io/blog/cloud-aws/https://justinqy.github.io/blog/cloud-aws/Operations in Amazon Web Services Type of Cloud Computing IaaS (Infrastructure as a Service) Amazon EC2, GCP(Google Cloud Platform), Azure... PaaS (Platform as a Service) Elastic BSun, 18 May 2025 13:13:43 GMTCloudhttps://justinqy.github.io/blog/ai-nlp-word-embeddings/https://justinqy.github.io/blog/ai-nlp-word-embeddings/Words to vector, catch word meanings by word embeddings. Representing words One-Hot Encoding Choose a vocabulary, each word in your sample is represented by a one-hot vector (shapeSun, 20 Apr 2025 17:27:28 GMTNLPhttps://justinqy.github.io/blog/ai-machine-learning-decision-tree/https://justinqy.github.io/blog/ai-machine-learning-decision-tree/Definition, implementation and pros and cons of decision tree models. What is a decision tree? A decision tree is a simple model for 'supervised' classification/regression. Each inFri, 11 Apr 2025 00:35:42 GMTMachine Learninghttps://justinqy.github.io/blog/ai-nlp-tvshow-script-generation/https://justinqy.github.io/blog/ai-nlp-tvshow-script-generation/Notes for NLP project Questions and Solutions 1. Llama2 Model takes more than 40G GPU RAM to train. Solution: LoRA: Low-Rank Adaptation of Large Language Models Concepts: Instead oThu, 27 Mar 2025 16:23:17 GMTNLPhttps://justinqy.github.io/blog/ai-nlp-bert-reading/https://justinqy.github.io/blog/ai-nlp-bert-reading/Record problems in the "BERT" paper. Process Analysis 1. masked language model: randomly mask some of the input tokens, and predict the originally vocab id of the mask ones. 2. nexWed, 19 Mar 2025 20:40:49 GMTNLPhttps://justinqy.github.io/blog/ai-nlp-sentiment-analysis-record/https://justinqy.github.io/blog/ai-nlp-sentiment-analysis-record/Recording a task to solve a movie comment sentiment analysis task Data Preprocessing Dataset downloading Download movie comments dataset (IMDb) from Hugging Face Data Cleaning PartTue, 18 Mar 2025 22:36:26 GMTNLPhttps://justinqy.github.io/blog/ai-models/https://justinqy.github.io/blog/ai-models/Steps to learn and master a kind of model 2W and 3H What is the model for? Like RNN Language Model is used for word prediction or words generation. How is the model built? Maybe soThu, 20 Feb 2025 18:04:56 GMTDeep Learninghttps://justinqy.github.io/blog/ai-machine-learning-plots-in-ml/https://justinqy.github.io/blog/ai-machine-learning-plots-in-ml/Some types of plots and how can they be helpful for machine learning tasks, such as feature selection. Data exploration During data exploration, we can use plots below to find moreSun, 09 Feb 2025 15:15:55 GMTMachine Learninghttps://justinqy.github.io/blog/ai-nlp-rnn/https://justinqy.github.io/blog/ai-nlp-rnn/Sequence Models and RNN Why use sequence models? Sequence models are used to situations when you have sequential input (e.g. images fo a human action, a paragraph of text or empty)Thu, 06 Feb 2025 10:05:06 GMTNLPhttps://justinqy.github.io/blog/ai-machine-learning-data-preprocess/https://justinqy.github.io/blog/ai-machine-learning-data-preprocess/Including methods implemented during EDA and data lacking. Lack of Training Data (数据短缺) --- Data Augmentation (数据增强) Expand an input dataset by slightly changing the existing (origMon, 03 Feb 2025 15:48:28 GMTMachine Learninghttps://justinqy.github.io/blog/ai-machine-learning-data-types/https://justinqy.github.io/blog/ai-machine-learning-data-types/Different data types in ML. Interval Data This is numerical data which has proper order and the exact zero . Temperature, time, credit score, pH... (zero value exists) Ratio Data TSun, 02 Feb 2025 20:39:10 GMTMachine Learninghttps://justinqy.github.io/blog/python-math-formula/https://justinqy.github.io/blog/python-math-formula/How to calculate mathematics in python 1. covariance np.cov(x1, x2) 2. correlationMon, 27 Jan 2025 10:47:41 GMTPythonhttps://justinqy.github.io/blog/python-tensorflow-pytorch/https://justinqy.github.io/blog/python-tensorflow-pytorch/Applications in tensorflow and pytorch. Tensorflow 1. transform "pandas.dataframe" to "tensorflow.dataset" For training dataset, you should set the "label" to be the prediction labMon, 27 Jan 2025 10:41:37 GMTPythonhttps://justinqy.github.io/blog/python-plotting/https://justinqy.github.io/blog/python-plotting/How to plot charts for dataset Plotting --- During a competition, there must be a number of features and some of them might make more influence on the target value than others. SoMon, 27 Jan 2025 10:41:06 GMTPythonhttps://justinqy.github.io/blog/python-pandas/https://justinqy.github.io/blog/python-pandas/Applications in pandas "pandas is kind of excel in python" 1. Dataframe 2. Dataframe to numpy: df.to numpy() 3. Numpy to Dataframe: df = pd.Dataframe(array) 4. drop some columns: 5Mon, 27 Jan 2025 10:40:54 GMTPythonhttps://justinqy.github.io/blog/python-numpy/https://justinqy.github.io/blog/python-numpy/Applications in numpy 1. reshape numpy array .reshape() 2. generate array from a fixed range 3. generate random values arrayMon, 27 Jan 2025 10:40:42 GMTPythonhttps://justinqy.github.io/blog/python-virtual-environment/https://justinqy.github.io/blog/python-virtual-environment/Python Venv 搭建个人python虚拟环境 --- 新建一个文件夹目录, 假设叫做“deeplearning”. 执行 python3 -m venv venv , 第一个“venv”表示创建virtual environment, 后面的“venv”表示环境文件夹命名. 执行 source venv/bin/activate 以激活python虚Mon, 27 Jan 2025 10:32:14 GMTPythonhttps://justinqy.github.io/blog/ai-machine-learning-math-in-ml/https://justinqy.github.io/blog/ai-machine-learning-math-in-ml/Mathematics in machine learning IQR (Interquartile Range) IQR describes the distance between the 1st quartile and the 3rd quartile. It is a method to detect outliers in dataset. ouSat, 25 Jan 2025 17:08:03 GMTMachine Learninghttps://justinqy.github.io/blog/ai-machine-learning-model-evaluation/https://justinqy.github.io/blog/ai-machine-learning-model-evaluation/Evaluate a model and decide what to do next. Model Evaluation Train Test Split Split the dataset to: 70% for training and 30% for testing. --- Train/test Procedure for Linear RegreFri, 17 Jan 2025 18:26:21 GMTMachine Learninghttps://justinqy.github.io/blog/ai-machine-learning-optimizing-gradient-descent/https://justinqy.github.io/blog/ai-machine-learning-optimizing-gradient-descent/Notions about gradient descent. Optimizing Learning Algorithm --- Feature Engineering (特征工程) Use intuition to create new features, by combining and transforming the original ones.Thu, 16 Jan 2025 21:18:33 GMTMachine Learninghttps://justinqy.github.io/blog/ai-cv-neural-style-transfer/https://justinqy.github.io/blog/ai-cv-neural-style-transfer/Neural Style Transfer in deep learning --- What is Neural Style Transfer Use a 'Content' image C and a 'Style' image S to generate a new image G , which has C's content and S's styMon, 13 Jan 2025 15:13:01 GMTCVhttps://justinqy.github.io/blog/ai-cv-face-recognition/https://justinqy.github.io/blog/ai-cv-face-recognition/1. How to solve face recognition problem with one-shot learning, which uses Siamese Network and Triplet loss function. 2. Face verification problem with binary classification, intrSat, 11 Jan 2025 20:49:50 GMTCVhttps://justinqy.github.io/blog/ai-cv-object-detection/https://justinqy.github.io/blog/ai-cv-object-detection/1. Algorithms for object detection. Object Detection Object detection contains 'Object Localization' and 'Landmark Detection'. --- Object Localization --- 1. Image classification (Fri, 03 Jan 2025 17:14:23 GMTCVhttps://justinqy.github.io/blog/ai-cv-image-classifier/https://justinqy.github.io/blog/ai-cv-image-classifier/1. Classic networks like LeNet-5, AlexNet and VGG; 2. Architectures like ResNet and Inception Net to improve performance of CNNs; 3. MobileNets to allow mobile devices to run appsFri, 27 Dec 2024 12:53:13 GMTCVhttps://justinqy.github.io/blog/ai-cv-convolutional-neural-network/https://justinqy.github.io/blog/ai-cv-convolutional-neural-network/Brief Introduction of Convolutional Neural Network, related architectures and computer vision practice --- Computer Vision Problems 1. Image Classification e.g. given an image of 6Sun, 15 Dec 2024 21:29:56 GMTCNNhttps://justinqy.github.io/blog/ai-cv-cnn/https://justinqy.github.io/blog/ai-cv-cnn/Delete after finish YOLO algorithm. 1. convolution (asterisk) 2. filter(kernel) 3. image filter = new image 4. python: conv forward, tensorflow: tf.nn.con2d, keras: conv2D 5. edgeMon, 09 Dec 2024 15:51:33 GMTCVhttps://justinqy.github.io/blog/python-terms-in-python-programming/https://justinqy.github.io/blog/python-terms-in-python-programming/-- 1. arguments: 实参, 调用函数时传入的参数 2. parameters: 形参, 定义函数时包含的参数 3.Wed, 04 Dec 2024 10:36:44 GMTPythonhttps://justinqy.github.io/blog/backend-fastapi/https://justinqy.github.io/blog/backend-fastapi/Recording the progress of deploying the NLP project by fastAPI, docker and streamlit. FastAPI Create an api for others to use to generate scripts by sending a prompt. 1. Create a nThu, 17 Oct 2024 15:17:29 GMTBackEndhttps://justinqy.github.io/blog/ios-objective-c/https://justinqy.github.io/blog/ios-objective-c/OC Objective-C 程序的编译过程 通过命令查看某个OC源文件编译过程: clang -ccc-print-phases xxx.m Clang前端处理 1. Lexical Analysis (词法分析) 2. Grammatical Analysis (语法分析) 3. Semantic Analysis (语义分析) 4. IntermediThu, 17 Oct 2024 15:17:29 GMTiOShttps://justinqy.github.io/blog/ios-ios-4/https://justinqy.github.io/blog/ios-ios-4/iOS客户端开发工程师面试题记录 Objective-C 语言特性： 1.面向对象中的封装, 继承, 多态分别是什么? 体现在哪些地方? 是怎么使用的? 2.动态类型（id类型）是什么? 在OC中是怎么使用的? 为什么需要使用id类型? 它和其他的类型有什么区别吗? 3.动态绑定（关键词@selector）是什么? 在OC中是如何使用的? 为什么需要使用动态Wed, 16 Oct 2024 17:57:53 GMTiOShttps://justinqy.github.io/blog/ai-machine-learning/https://justinqy.github.io/blog/ai-machine-learning/Todo List 1. Neural Network Notes 2. K-neighbors and SVM method 3. Relu function 3. multiclass classification (softmax) and numerical round-off error (using linear output and 'fromWed, 24 Jul 2024 17:24:50 GMTTodohttps://justinqy.github.io/blog/ai-deep-learning-neural-network-neural-network-model/https://justinqy.github.io/blog/ai-deep-learning-neural-network-neural-network-model/Neural Network Model Why do we need Neural Network? --- There is a weakness in both linear regression and logistic regression: the model needs to do a large amount of calculation wSat, 13 Jul 2024 17:04:51 GMTDeep Learninghttps://justinqy.github.io/blog/python-scikit-learn/https://justinqy.github.io/blog/python-scikit-learn/记录scikit learn库的使用 Logistic Regression --- Loading Dataset Split the train/test set Fit the Model Make Predictions Evaluate AccuracyFri, 12 Jul 2024 15:30:45 GMTPythonhttps://justinqy.github.io/blog/ai-machine-learning-logistic-regression/https://justinqy.github.io/blog/ai-machine-learning-logistic-regression/logistic regression model Logistic Regression --- Compared with linear regression, logistic regression is used to solve questions which only have limited possible answers. For examThu, 11 Jul 2024 10:41:08 GMTMachine Learninghttps://justinqy.github.io/blog/ai-kaggle-kaggle-2/https://justinqy.github.io/blog/ai-kaggle-kaggle-2/Kaggle小白快速上手指北 准备工作 --- 比赛选择 上来先筛选出新手模块的competition, 因为我本人是新手, 并且想练习课程中学到的知识, 所以就从简单的开始了 了解比赛信息 Description 描述信息 根据"Description"模块可以快速了解题目背景, 因为kaggle中的题目(我目前接触到的)都与现实紧密结合, 所以我认为多了Mon, 08 Jul 2024 21:18:25 GMTKagglehttps://justinqy.github.io/blog/ai-machine-learning-linear-regression/https://justinqy.github.io/blog/ai-machine-learning-linear-regression/linear regression model 线性回归模型(Linear Regression) --- 线性回归模型可以根据一些特征值(feature x)数据, 计算一个可能的结果(预测), 这个结果是无限多个数中的一个. 例如根据房屋个数, 房屋面积, 花园面积和房屋建成年份等特征值进行房价预测等. In a word, linear regressMon, 08 Jul 2024 13:36:00 GMTMachine Learninghttps://justinqy.github.io/blog/ai-kaggle-kaggle/https://justinqy.github.io/blog/ai-kaggle-kaggle/Titanic - Machine Learning from Disaster --- 预测泰坦尼克游客是否能够幸存 题目地址 House Prices - Advanced Regression Techniques --- 预测房屋售价 题目地址Sat, 06 Jul 2024 15:52:21 GMTKagglehttps://justinqy.github.io/blog/ios-ios-objectivec/https://justinqy.github.io/blog/ios-ios-objectivec/记录一些Objective-C的使用技巧和注意事项, 也算是对iOS开发生涯的总结 (in progress) 面向对象思想 --- 面向对象和面向过程的区别 1. 面向过程强调功能行为，关注解决问题需要哪些步骤 (所有过程亲力亲为) 2. 面向对象将功能封装进对象，强调具备了功能的对象，关注的是解决问题需要哪些对象 (找到具有对应功能的对象，让对象去做事情Fri, 05 Jul 2024 02:40:49 GMTiOSObjective-Chttps://justinqy.github.io/blog/ios-ios-3/https://justinqy.github.io/blog/ios-ios-3/记录iOS多线程开发的基本知识和使用方法, 积累iOS开发经验 一些基本概念 --- 一个应用程序可以对应多个进程, 每个进程中至少有一个线程, 进程中的线程共享该进程的资源. 线程执行任务的方式 -- 串行（任务和任务之间有执行顺序，即多个任务一个一个地按顺序执行，一个线程同时只能执行一个任务） 单个进程中的每条线程可以并行执行任务 同一时间CPU只能处理Fri, 05 Jul 2024 02:40:48 GMTiOS多线程https://justinqy.github.io/blog/ios-ios/https://justinqy.github.io/blog/ios-ios/单例模式记录 单例模式提供了一种创建对象的最佳方式。这种模式涉及到一个单一的类，该类负责创建自己的对象，同时确保只有单个对象被创建。 单例模式的作用 --- 保证在程序运行过程中，一个类只有一个实例，该实例易于供外界访问，从而方便地控制了实例个数，节约系统资源 单例模式的使用场合 --- 在整个应用程序中，需要共享一份资源（资源只需要初始化一次） 单例模式的Fri, 05 Jul 2024 02:40:47 GMTiOS设计模式https://justinqy.github.io/blog/ios-ios-2/https://justinqy.github.io/blog/ios-ios-2/观察者模式记录 什么是观察者模式 --- 1. 观察者模式定义：观察者模式定义了一种一对多的依赖关系，让多个观察者对象同时监听某一个主题对象。这个主题对象在状态上发生变化时，会通知所有观察者对象，使它们能够自动更新自己。 2. iOS中实现观察者模式：Notification、KVO。 Notification -- 通知 --- 现有对象A和B，A对B的变Fri, 05 Jul 2024 02:40:47 GMTiOS设计模式https://justinqy.github.io/blog/markdown-markdown/https://justinqy.github.io/blog/markdown-markdown/总结了常用的markdown语法, 方便平常整理笔记时使用 标题 --- 使用 号表示标题，支持 1 到 6 级标题。 段落 --- 段落之间需要空一行。 强调 --- 使用 或 表示斜体，使用 或 表示粗体。 删除线 --- 使用 表示删除线。 引用 --- 使用 表示引用。 列表 --- 无序列表 使用 -、 或 + 表示无序列表。 有序列表 使用数字加Thu, 04 Jul 2024 22:40:50 GMThttps://justinqy.github.io/blog/ios-ios-datarace/https://justinqy.github.io/blog/ios-ios-datarace/多线程数据竞争demo 数据竞争代码举例以及用互斥锁解决Thu, 04 Jul 2024 04:00:00 GMTiOS多线程https://justinqy.github.io/blog/ios-ios-deadlock/https://justinqy.github.io/blog/ios-ios-deadlock/多线程死锁理解笔记(乱) 使用disptach sync在串行队列(包括主队列)导致的死锁问题原理解释:Thu, 04 Jul 2024 04:00:00 GMTiOS多线程https://justinqy.github.io/blog/ios-ios-gcd/https://justinqy.github.io/blog/ios-ios-gcd/多线程之gcd记录 GCD (grand central dispatch) 概念 gcd是一个强大的中枢调度器, 基于C语言实现, 解决多核的并行运算，能够自动利用更多的CPU内核，自动管理线程的生命周期. 任务: 执行什么操作(可以理解为函数的内容) 队列: 用来存放、安排任务. 不同的队列类型有不同的执行任务的策略.(串行队列在当前线程按顺序执行任务,Thu, 04 Jul 2024 04:00:00 GMTiOS多线程https://justinqy.github.io/blog/ios-ios-mjextension/https://justinqy.github.io/blog/ios-ios-mjextension/MJExtension MJExtension -- Json转Model 一. Json数据 定义： JSON(JavaScript Object Notation) 是一种轻量级的数据交换格式。JSON采用完全独立于语言的文本格式，这些特性使JSON成为理想的数据交换语言。易于人阅读和编写，同时也易于机器解析和生成。 ----- Json可以将js对象中Thu, 04 Jul 2024 04:00:00 GMTiOS第三方库https://justinqy.github.io/blog/ios-ios-mrc-arc/https://justinqy.github.io/blog/ios-ios-mrc-arc/MRC & ARC 内存管理模型 一. 需要进行内存管理的对象 1. 任何继承了NSObject的对象需要进行内存管理 2. 非对象类型(int、char、float、double、struct、enum等) 不需要进行内存管理 二. 内存结构 1. 堆 一般由程序员分配释放，若程序员不释放，程序结束时 可能 由OS回收，分配方式类似于 链表 ，继承了NSOThu, 04 Jul 2024 04:00:00 GMTiOS内存管理https://justinqy.github.io/blog/ios-ios-nsthread/https://justinqy.github.io/blog/ios-ios-nsthread/NSthread记录 创建子线程 --- 子线程状态切换 --- 线程间通信实现 ---Thu, 04 Jul 2024 04:00:00 GMTiOS多线程https://justinqy.github.io/blog/ios-ios-pthread/https://justinqy.github.io/blog/ios-ios-pthread/多线程pthread记录Thu, 04 Jul 2024 04:00:00 GMTiOS多线程https://justinqy.github.io/blog/ios-ios-ui/https://justinqy.github.io/blog/ios-ios-ui/UI控件 UI控件 UIView 视图类 用法： UIView view = [[UIView alloc] init]; 常见属性 1. 获得自己的父控件对象 @property (nonatomic, readonly) UIView superview; //父控件只有一个 2. 获得自己的子控件对象 @property (nonatomic, reaThu, 04 Jul 2024 04:00:00 GMTiOSObjective-Chttps://justinqy.github.io/blog/ios-ios-uicollectionview/https://justinqy.github.io/blog/ios-ios-uicollectionview/UICollectionView 学习UICollectionView UICollectionView的组成 1. Cell -- 用于展示内容，尺寸和内容可以各不相同。 2. Supplementary Views -- 追加视图，类似于UITableView中每个Section的Header或Footer 3. Decoration View -- 装Thu, 04 Jul 2024 04:00:00 GMTiOSObjective-Chttps://justinqy.github.io/blog/python-basic/https://justinqy.github.io/blog/python-basic/Python basic knowledge 列表创建 1. 创建一个 一维 空列表, 长度为0 2. 创建一个 二维 空列表, 行和列长度为 n 列表的几种遍历方式: 1. 从0到nums列表最后一位顺序遍历法, i表示下标 ( nums[0], nums[1]...nums[n-1] ) 2. 遍历由 [left, right] 左右闭区间组成的numsWed, 03 Jul 2024 15:49:26 GMTPython