Ai learn
1.0.0
Created by Tang Yudi
The original intention of organizing this project is to facilitate students to quickly start the artificial intelligence self-study plan, avoid detours in the learning process, and get started with AI as quickly as possible and start practical projects. It provides nearly 200 AI practical cases and projects . These are not online What I have collected are cases developed and accumulated by me in the past five years of online and offline teaching. It can be said that they have been updated repeatedly and iteratively, and are suitable for students to learn and practice step by step. Students who come here, remember to click a star to save it!
At the end of 2019, I published the supporting textbook for the machine learning course "Learn Python Data Analysis and Machine Learning Practice with Diego". The style is still easy to understand. It took two years and more than ten revisions to finally meet you. In order to facilitate more students to start their study plan quickly, I decided to give you the electronic version of this book for free . I hope it can bring learning gains to everyone! You can download the PDF version on the homepage of this project. If you like the teaching materials, you can also purchase them from JD.com.
"Learn Python data analysis and machine learning practice with Diego" PDF original download :
(Network disk link: https://pan.baidu.com/s/19wzJeyPmwTBDp9ASEWBvFQ Extraction code: tece )
The following catalog is the learning route. Beginners are advised to study in the order given in the catalog. Students who have already started can choose according to their own preferences.
The data involved in the case are all real data sets, and some will be quite large. Directly uploading them to github will be very slow for everyone to download. I will gradually upload the network disk links of each module, which include data, code, PPT and other learning resources. If you need supporting video explanation, please add WeChat: digexiaozhushou (Pinyin of Di Ge’s Little Assistant)
If you have any cooperation, communication or project issues in various aspects, you can directly add WeChat: digexiaozhushou (Pinyin of Dige Assistant)
To learn artificial intelligence (data science), you still need some basic skills. The most basic and core ones are Python and mathematics ! It’s not difficult for these two brothers to get started. It’s okay to master the basics first and learn while using them!
If you are not familiar with Python, I suggest you watch my introductory Python video course to get started quickly! portal
The most direct explanation is that everyone uses it! It used to be object-oriented programming, but later people preferred copy-and-paste programming, and now they are too lazy to program for GitHub. It is true, you have to be lazy when you should be lazy, and Python does this! All subsequent practical content will be based on Python, so you have no choice!
Anaconda is enough! Anaconda is enough! Anaconda is enough! Okay, I’ve said it three times. For a detailed explanation, just refer to the portal course above.
The toolkit means that others have written all the functions, and we can call it directly and that’s it! There are corresponding tool kits for data processing, analysis, modeling, etc. For learning, there is no need to memorize these tool kits. You need to familiarize yourself with them first, and you will definitely need to use them and check them later.
| Toolkit name | Function overview |
|---|---|
| Numpy | A must for matrix calculations! It is the core of all subsequent calculations and the core toolkit in the field of data science. |
| Pandas | A must for data processing! Reading data, processing data, and analyzing data must be done by him! |
| Matplotlib | Visualization is a must! The function is very powerful. There is no picture that cannot be drawn. Analysis and display rely on it! |
| Seaborn | A simpler visualization tool! One line of code gives you a visual display of the results |
Students must be very aware of how important mathematics is, especially in the field of artificial intelligence (data science). It is difficult to do anything without knowing mathematics. Many students have asked me a question, can so much mathematics be really used in work? Let me explain to you that the artificial intelligence industry is developing very rapidly. In actual work, you must learn while doing it. What should you learn? It must be some of the outstanding papers currently. If you can't even understand basic mathematical formulas, then there is no need to talk about high-end technology. Students in this field will definitely have this idea. The so-called artificial intelligence is just doing various mathematical calculations on data!
I don’t think you need to start from scratch and spend a lot of time learning step by step. For example, my colleagues and friends and I have been doing this for a long time. I don’t know how many times I have studied mathematics. I have solved countless questions at times, but I will also encounter this problem. Many knowledge points will be forgotten quickly if I haven't looked at them for a while. One of the things I do most often is to search for whatever I use. The search process is actually a process of learning and progress. It is recommended that you quickly go through common knowledge points (advanced mathematics, linearity, basics of probability theory). During this process, do not look at various problem-solving processes, nor do you need to worry about the specific solution methods. To put it bluntly, just understand What a formula does and what it is used for is enough. Similar to the exercises in the textbook and the solutions in the exercise book, you don’t need them. In the future, you won’t use a pen to calculate these troublesome things. Save this time to study. Algorithms are more cost-effective!
| Knowledge points | content | effect |
|---|---|---|
| Advanced Mathematics | Basics of advanced mathematics, calculus, Taylor's formula and Lagrange, | Essential for deriving machine learning formulas |
| linear algebra | Basics of linear algebra, eigenvalues and matrix decomposition, | Necessary for algorithm solution |
| probability theory | Basics of probability theory, random variables and probability estimation, commonly used distributions | Machine learning often mentions these words |
| Statistical analysis | Regression analysis, hypothesis testing, correlation analysis, analysis of variance | Essential for data analysis |
The core of the field of artificial intelligence is machine learning. No matter which direction you want to pursue in the future, you must start with machine learning! There are two main things. The first is to master the principles of classic algorithms, and the second is to be proficient in using the Python toolkit for practical modeling!
What to learn about algorithms? Understand how machine learning algorithms operate on data to complete the modeling and solution process. To put it bluntly, it means becoming familiar with how mathematics is used in algorithms. The important thing is to understand! Don't dwell on a problem endlessly, it's a waste of time, and you might be able to solve it in no time during the subsequent learning process. I think you must study algorithms more than once, especially for students who are preparing for job interviews. It is normal to study the algorithm twice or three times (a classmate once told me that he studied the course a total of 6 times before the interview)
With deep learning, is machine learning still needed?
Deep learning can be said to be a type of machine learning algorithm. It does not mean that other classic algorithms are not needed with neural networks. The most appropriate algorithm needs to be selected based on different tasks and data. The learning path must start with machine learning. In fact, it is really easy to look at neural networks after mastering these classic algorithms!
| Knowledge points | content | Overview |
|---|---|---|
| Classification algorithm | Logistic regression, decision tree, support vector machine, ensemble algorithm, Bayesian algorithm | Students preparing for interviews must master |
| Regression algorithm | Linear regression, decision trees, ensemble algorithms | Some algorithms can do both classification and regression |
| Clustering algorithm | k-means, dbscan, etc. | Unsupervised is only considered when there is really no label. |
| Dimensionality reduction algorithm | Principal component analysis, linear discriminant analysis, etc. | Focus on understanding the idea of dimensionality reduction |
| Advanced algorithms | GBDT boosting algorithm, lightgbm, EM algorithm, hidden Markov model | Students who have time and energy can challenge advanced algorithms |
Analyze the impact of classic algorithm modeling methods and parameters on results through comparative experiments, and understand the parameters and application examples in the algorithms through experiments and visual displays.
| Case name | Content overview |
|---|---|
| Linear regression experimental analysis | Master univariate and multiple linear regression, nonlinear regression methods, and the role of regularization penalty |
| Model evaluation method | Comparison of commonly used classification and regression algorithm evaluation methods, examples of data set segmentation |
| Logistic regression experimental analysis | Classic classification model construction method, decision tree boundary drawing method |
| Clustering algorithm experimental analysis | Unsupervised modeling examples, clustering algorithm evaluation methods, unsupervised role and application examples |
| Decision tree experimental analysis | Tree model visualization examples and construction methods, classification and regression applications of tree models |
| Integrated algorithm experimental analysis | Application examples and effect analysis of integration methods, comparison of common integration strategies |
| Support vector machine experimental analysis | SVM involves parameters and modeling comparison experiments |
| Practical analysis of association rules | Essential knowledge points and modeling analysis examples of association rules |
In order to better understand the mechanism of the algorithm, we reproduce the classic algorithm from scratch, adhere to the principle of no packet loss, and complete all the modules required by the algorithm step by step.
The main purpose is to better grasp the working principle of the algorithm, and the emphasis is on practice! Students who have time can reproduce it themselves, but students who are short on time do not have to do it themselves.
| Case name | Content overview |
|---|---|
| Linear regression code implementation | Commonly used functions of module building algorithms |
| Logistic regression code implementation | Example interpretation of logistic regression implementation method |
| Kmeans code implementation | Very simple and easy to understand unsupervised algorithm |
| Decision tree code implementation | The tree model is actually a recursive implementation |
| Neural network code implementation | The amount of code is slightly large, so it is recommended to learn debug mode. |
| Bayesian code implementation | Bayes is still easier to explain in text tasks |
| Association rule code implementation | Commonly used data analysis algorithms |
| Build a music recommendation system | Construct a recommendation system model from scratch |
In actual combat, the mathematical knowledge points may be weakened, because most of the time we use ready-made tool kits to complete tasks (package transfer man). There are many energy-saving functions that everyone needs to master here. The first is to be proficient in using these common toolkits. Data preprocessing, feature engineering, parameter adjustment, and verification are all very core steps. In summary, the processes and routines required to complete different tasks are similar, but the methods and algorithms used may be different, which requires everyone to continuously accumulate to enrich practical experience. These cases provided to students can be used as their own practical templates!
The most important thing is to learn how to preprocess and analyze different data (numeric values, text, images), skillfully apply the major core functions in the toolkit to participate in preprocessing, propose multiple solutions for different tasks, and conduct experimental analysis. To sum up, do more experiments and do more hands-on work. The more you write code, the more proficient you will become!
| Case name | Content overview |
|---|---|
| K nearest neighbor algorithm in practice | Machine learning introductory case, mastering the toolkit applied to modeling methods |
| Transaction data anomaly detection | Very important, detailed analysis and comparison of data processing and modeling strategies |
| Integrated algorithm modeling practice | I don’t need to say more about integration, it’s a must-have core strategy. |
| Temperature prediction based on random forest | Random forest is the most commonly used algorithm in machine learning. Detailed analysis and comparison |
| News classification practice | Text data analysis and processing, and practical modeling based on Bayesian algorithm |
| Clustering practice analysis | Unsupervised application examples |
| time series analysis | Time series data production method, modeling based on sequence data |
| User churn warning | I often say that Fantasy Westward Journey users are losing, this is just a DEMO |
| Use lightgbm for hotel traffic prediction | Another big killer, even worse than xgboost |
| Census Dataset Project Practice-Income Forecast | Core templates, data analysis, visualization, etc. are all there. |
| Bayesian Optimization in Practice | More difficult, Bayesian optimization toolkit usage examples |
| Comparison of text feature methods | Comparison of commonly used feature extraction methods for text data |
| Make your own tool kit | Make your own bag for fun |
| Case name | Content overview |
|---|---|
| Python practical association rules | It’s so easy to use the toolkit to make association rules |
| Airbnb Dataset Analysis and Modeling | Housing price data set analysis and modeling examples |
| Hotel recommendation system based on similarity | To build a recommendation system to complete hotel recommendations |
| Product sales regression analysis | Sales forecast, a very routine task, can be done with routine methods |
| PUBG data set exploration analysis and modeling | PlayerUnknown's Battlegrounds data set, let's see who killed you |
| Model interpretation method in practice | How to interpret the model after modeling? These toolkits can help you do it. |
| Essential toolkit for natural language processing in practice | Interpretation of common NLP toolkits and practical examples |
| Bank customer repayment probability forecast | Bank customer data to predict likelihood of repayment |
| Image feature cluster analysis practice | How to cluster image data? |
| Census Dataset Project Practice-Income Forecast | Core templates, data analysis, visualization, etc. are all there. |
Everyone hears the word data analysis every day, but what should we do? It is nothing more than obtaining valuable information from data, and there are still many methods and routines. This direction does not require any theoretical accumulation, just use the data and do it! The accumulation of cases is the learning process!
Simply put, data mining is the application of machine learning algorithms to massive amounts of data to obtain the desired results. The focus in data mining is not the choice of machine learning algorithm, but how to process the data to get better prediction results. Here feature engineering and preprocessing will become the core solution.
| Case name | Content overview |
|---|---|
| Titanic rescue predictions | Classic kaggle competition case, the first practical project for entry-level data mining |
| Data feature construction | Feature engineering is the core of data mining. Based on sklearn, various feature construction methods are explained. |
| User portrait practice | Everyone must have heard of user portraits. How to apply data to complete the portrait? |
| Integration strategy example | In data mining, integration strategies are usually chosen to better improve the effect. |
| Xgboost actual combat | A typical representative in integration and a killer weapon in the competition |
| JD.com purchase intention forecast | Classic prediction problem, complete prediction tasks based on user historical behavior data |
| kaggle data science survey | Visually display the participants in the kaggle competition |
| house price forecast | Data mining entry-level cases to quickly master the routines |
| Power Sensitive User Analysis | Competition examples, mainly explaining the role of feature engineering |
| fbprophet time series prediction | A very practical algorithm for time series forecasting, very simple to use |
We have selected large-scale competition cases such as Tianchi, Kaggle, and Rongji for everyone, and the codes and solutions provided are the solution ideas of the winners during the competition. Just like if you want to learn to play chess, you have to play with the best players to improve yourself. Each case will explain the winner's ideas and overall solution and provide code implementation. Very helpful for everyone to improve!
| Case name | Content overview |
|---|---|
| Kuaishou short video user activity analysis | Predict next activity based on user behavior data |
| Industrial Chemical Production Forecast | Analyze chemical industry data and model to predict production efficiency |
| Smart city-road travel time prediction | A very down-to-earth competition, predicting travel time based on road data |
| Feature Engineering Modeling Interpretable Toolkit | One of the most difficult aspects of data mining is feature interpretation. These toolkits are very useful. |
| Medical diabetes data named entity recognition | Named entity recognition algorithm explanation and application example analysis |
| Loan platform risk control model-feature engineering | Using graphical models to build feature engineering is a widely used idea. |
| News keyword extraction model | Keyword extraction can be said to be an essential skill for NLP |
| Machine learning project practical template | The template is here. You can apply it to future tasks. The methods are similar. |
| Power Sensitive User Analysis | Competition examples, mainly explaining the role of feature engineering |
Data mining mainly uses modeling to make predictions, while data analysis focuses on visual display and analysis of the impact of various indicators on the results. We have selected some classic analysis cases for you, many of which can be used directly as templates.
| Case name | Content overview |
|---|---|
| Scatter plot drawing techniques | It has been said that visualization is the key point, and drawing is definitely a must. |
| New York Taxi Operation Analysis and Modeling | Having used many toolkits, you can become familiar with how to analyze and display geographic data. |
| Movie recommendation task based on statistical analysis | Commonly used methods for statistical analysis and recommendations |
| Data Analysis and Machine Learning Templates | This template is really comprehensive, including analysis, display, modeling, and evaluation. |
| Data dimensionality reduction | Comparative analysis and display of several commonly used dimensionality reduction algorithms |
| Product visual display and text processing | Text data preprocessing and visual display |
| multivariate analysis | Multivariate analysis is also a common method in data analysis |
| Product order data set analysis | Order data set analysis |
| KIVA loan data analysis | Loan data set analysis |
Deep learning can be said to be the most useful algorithm at the moment, and it can be used in various fields. In fact, the core is still in computer vision and natural language processing, because neural network algorithms are more suitable for image and text data. The main things you need to master are algorithms and frameworks. Algorithms are classic network models such as CNN and RNN, and frameworks are practical tools such as tenorflow, Pytorch, etc., which will be discussed in detail later.
It seems that when many friends get a task now, their first thought is to use deep learning directly. If deep learning was difficult and cumbersome to do, would it still be so popular? In fact, on the contrary, I think deep learning is really much simpler than machine learning. In machine learning, we need to choose different preprocessing methods and feature engineering construction methods for different data. The routines in deep learning are relatively more fixed, and with these open source frameworks and major classic network architectures, all we usually need to do is apply them. The overall difficulty is easier than the machine learning task (relatively speaking!).
| Algorithm name | Content overview |
|---|---|
| neural network | Neural network is the most basic, which is equivalent to laying the foundation for the learning of subsequent networks. |
| convolutional neural network | This sounds familiar to everyone, the big brother in deep learning! Computer Vision Core Network |
| recurrent neural network | Bei Qiao Feng and Nan Murong are the big brothers in natural language processing! |
| adversarial generative network | It is a popular model now. It is fun to play with and can be used for various image fusions. |
| sequence network model | Commonly used architectures in NLP, machine learning translation models, have many application points |
| Major classic network architectures | The CNN and RNN mentioned just now are relatively basic network models, and there are many extensions based on them that everyone needs to master. |
The framework is like saying that you design a network model, but it would be too troublesome to complete all the specific calculation processes yourself. The framework provides an efficient calculation method and does not require us to complete it, a set of fully automatic calculations. It is equivalent to that we only need to design the structure and leave the specific construction to it. What is necessary to play deep learning is a framework.
Tensorflow, Pytorch, keras, caffe, etc., there are so many frameworks, which one should I choose? Are there big differences between different frameworks? The most mainstream ones now are tensorflow and PyTorch, which are equivalent to KFC and McDonald's. Both are very strong. As for which one to choose, please refer to your respective project team and task requirements. If I had to recommend one, I would recommend PyTorch to everyone because it is more concise and popular. I have used all of these frameworks. The main reason is that I often need to refer to papers and open source projects in my work. I usually follow up on the frameworks used for the source code in other people's papers for secondary development, so everyone will use these frameworks sooner or later!
There is no theory to talk about the framework, and there is no need to read all kinds of long explanations, just use it! In fact, it is a toolkit, you can learn while using it, and the cases can be used as templates to summarize!
| Frame name | Content overview |
|---|---|
| Caffe framework | The god-level framework of ancient times has fallen into disgrace now. It was the first framework I learned. |
| Tensorflow2 version | Version 2 has made many improvements and is finally more user-friendly. It is much more comfortable to use than version 1. |
| Keras | In one sentence, it’s simple! Simple! Simple! No need to learn, it’s very easy to understand by looking at the code |
| PyTorch | The most popular framework at this stage, I estimate it will also be the most popular framework this year (2020), recommended! |
We provide you with a wealth of practical cases for each major deep learning framework. Which one you use depends on your preference!
I don’t need to explain that it is produced by Google. So many of Google’s open source projects are definitely based on the TF framework. If you want to study or refer to other people’s open source projects and papers, you must learn TF. It is also widely used in the industry. This wave will definitely pay off!
| Case name | Content overview |
|---|---|
| tensorflow installation and introduction | Version 2 is introduced in the installation method, just go through it briefly. |
| Interpretation of neural network principles and overall architecture | Review neural network architecture |
| Build a neural network for classification and regression tasks | Use TF to complete basic classification and regression tasks and master its application methods |
| Convolutional neural network principles and parameter interpretation | The architecture of CNN is explained in detail for each parameter. |
| Cat and dog identification practice | The classic image classification task, there is a lot to talk about here, it is very important |
| Image data enhancement examples | It can be said that data enhancement is now a must-have skill. |
| Training Strategy-Transfer Learning in Practice | The effect brought by transfer learning is still quite good. |
| Interpretation of the principles of recursive neural networks and word vectors | RNN model interpretation |
| Implementing word2vec based on TensorFlow | Word vector model interpretation and implementation based on TF |
| Text classification task based on RNN model | Complete text classification tasks based on TF |
| tfrecord creates data source | Data source production example |
| Apply CNN network to text classification practice | CNN can also perform text classification |
| time series forecasting | Time series data processing and modeling examples |
| Adversarial Generative Networks in Practice | GAN is here, this is so fun |
| Practical image fusion based on CycleGan open source project | My favorite GAN to play, the effect is quite amusing! |
| Classic network architecture Resnet in practice | You can learn the network architecture you must understand! |
At the end of 2019, the number of users of the Pytorch framework has surpassed tensorflow to become the most popular framework at the moment. The reason is actually very simple. Everyone likes to use simpler and easier-to-understand frameworks. The overall feeling is indeed easier to use than tensorflow and very convenient to debug. It is also recommended that beginners give priority to the Pytorch framework.
| Case name | Content overview |
|---|---|
| Basic processing operations of the PyTorch framework | Just get familiar with PyTorch and it’s very easy to get started. |
| Neural network practical classification and regression tasks | Using PyTorch to build a neural network model is indeed easier to use than TF |
| Convolutional neural network principles and parameter interpretation | CNN model architecture and parameter book interpretation |
| Practical interpretation of image recognition core module | Very important, the image processing core module in PyTorch |
| The role and application examples of transfer learning | Loading models in PyTorch for transfer learning |
| Interpretation of the principles of recursive neural networks and word vectors | Interpretation of RNN model architecture |
| Practical practice on text classification of news data sets | Build a text classification model based on PyTorch |
| Principles and practical analysis of adversarial generative network architecture | Popular interpretation of GAN model |
| Practical image fusion based on CycleGan open source project | PyTorch version of CYCLEGAN, this open source project is very well written |
| OCR text recognition principle | The principle of OCR is actually very simple and requires the assistance of multiple models to complete it. |
| OCR text recognition project practice | Build OCR network model |
| Video analysis and action recognition based on 3D convolution | Use 3D convolution to process video data and complete behavior recognition |
| Practical BERT model based on PyTorch | The architecture of BERT is so popular, it is one of the must-have models. |
| Interpretation of practical templates of PyTorch framework | Provide a template so that future tasks can be improved based on the template |
The overall feeling is that you don’t need to learn anything, you can just use it directly from the case. The TF2 version is actually very similar to keras. Suitable for doing experiments and writing papers, simple and fast!
| Case name | Content overview |
|---|---|
| Installation and introduction | Keras is easy to install and get started, based on tf |
| Build a neural network model | Build a neural network model to test the waters |
| Another battle with convolutional neural networks | CNN models are also very easy to build |
| LSTM time series forecasting task | LSTM model applied to time series tasks |
| Text classification practice | Text classification example |
| Multiple labels and multiple outputs | Multi-label tasks are very common and have great learning value |
| Practical practice on text classification of news data sets | Text classification task based on keras |
| data augmentation | Interpretation of data augmentation examples |
| adversarial generative network | GAN architecture, it is easier to use keras |
| Transfer learning and Resnet residual network | You must play with the resnet model yourself |
| Address Postcode Multiple Sequence Task | Text model example |
| seq2seq network practice | The sequence network model is still widely used. |
| Summary of practical templates | The keras template is provided for everyone. If you have any tasks, just write them directly. |
I think tensorflow and pytorch are already available at this stage, and it is not caffe’s turn to appear for the time being. I would not recommend it to beginners. There may be some papers and tasks that still require the caffe framework. Students who need it can just get it themselves!
| Case name | Content overview |
|---|---|
| Caffe configuration file interpretation | Interpretation of commonly used configuration files of Caffe framework |
| Various data set construction methods | Data set construction method, this is very important |
| Interpretation of common Caffe tools | There are many small tools built into it to complete tasks quickly |
| Face detection in practice | Build a face detection model based on the Caffe framework |
| Practical practice of locating facial key points | Complete facial key point recognition model based on Caffe framework |
I don’t need to say much about the computer vision industry, it is the most popular one right now. So what do you need to learn? The core is actually two parts, one is image processing and the other is image modeling. The so-called image processing is what Opencv does. This toolkit is simply invincible. You can find everything you want to use here. Image modeling mainly uses deep learning to complete tasks such as detection and recognition. At this stage of study, I think you don’t need to read about traditional image processing algorithms. You can simply familiarize yourself with them. The mainstream direction is still to use deep learning, which requires everyone to read more latest papers.
It is recommended that you choose the Python version to learn and use it. Just like other toolkits, just adjust it and you’re done! If you encounter something you are not familiar with, check the API more. Learning while using it is the fastest way. Basically all the functions in Opencv involve a lot of mathematical formulas. You can put these aside first. If you learn every algorithm and every formula, it will take you years and years. It will be completely timeless to use them in the future.
We have prepared a lot of learning resources and cases for everyone. In the early stage, you only need to be familiar with it. The toolkit is used, and copy-paste programming is also a skill!
| Case name | Content overview |
|---|---|
| Opencv introduction and environment configuration | Environment installation and configuration |
| Basic image operations | Use opencv to complete basic image processing operations and practice! |
| Thresholding and Smoothing | The most commonly used processing operations can be completed with just a few lines of code |
| Image morphology operations | Just get familiar with these morphological operations |
| Image gradient calculation | Image gradient calculation example |
| edge detection | Edge detection has a wide range of applications |
| Image Pyramid and Contour Detection | Contour detection example, the effect is still good |
| Histogram and Fourier Transform | Just get familiar with it |
| Project Practice-Credit Card Digital Identification | Do a practical project to detect and identify credit card numbers |
| Project Practice-Document Scanning OCR Recognition | Scan document data for OCR recognition |
| Image Features-harris | Commonly used feature extraction methods, the algorithm is simple and familiar |
| Image features-sift | The oldest feature extraction method, it still requires a lot of mathematics. |
| Case practice-panoramic image stitching | Everyone must have played with panoramic camera, how to achieve it? |
| Project Practice-Parking Space Identification | Heavy-duty project, building a parking space recognition model from scratch |
| Project Practice-Answer Card Identification and Judging | Let’s also have fun with automatic marking |
| Background modeling | Conventional treatment methods |
| Optical flow estimation | Just be familiar with it |
| Opencv DNN module | Load the trained model for recognition |
| Project Practice - Goal Tracking | The tracking effect is quite interesting. |
| Convolution principles and operations | Wherever the convolution goes, it is the core. |
| Project practice-fatigue detection | Detecting fatigue based on camera |
I recommend that students who are preparing for job interviews read through it all. The ideas in it are quite good, and most of them are reproduced based on papers. Students who have time are best to read the papers before starting to study the code. The code inside The amount will be relatively large, so it is recommended to start with the debug mode and look at the code line by line. During the explanation process, I will also enter the debug mode to explain it line by line.
We highly recommend the Mask-rcnn practical project. It can be said to be a universal project in computer vision. It can detect, identify, and segment in one step! The application scenarios are very wide, and it is also suitable for secondary development and improvement. If you want to write it on your resume, it must be it. Everyone needs to be familiar with the algorithm principle and source code. In the course, I will focus on explaining the project and apply it to myself. The data task is in progress!
| Project name | Content overview |
|---|---|
| Image style transfer (style-transfer) | Mainly to learn his ideas, the effect is still very interesting |
| Automatically complete missing images | There are many application scenarios for GAN network, and images can also be repaired by themselves. |
| Super-resolution reconstruction | One of the key areas of research in recent years, the results of this paper are already very good. |
| Object detection framework-MaskRcnn project | This is the open source project I highlight, a must-see! A must see! A must see! |
| Detailed explanation of MaskRcnn network framework source code | The source code is very important, you need to understand every line! |
| Train your own data based on the MASK-RCNN framework | How to label image data and train it? Here is your answer |
| Human posture recognition demo | There are many application scenarios for MaskRcnn |
| Object detection FasterRcnn series | A classic work on object detection that can be used as a learning resource |
| Practical image fusion based on CycleGan open source project | PyTorch version of CYCLEGAN, this open source project is very well written |
| OCR text recognition principle | The principle of OCR is actually very simple and requires the assistance of multiple models to complete it. |
| OCR text recognition project practice | Build OCR network model |
| Video analysis and action recognition based on 3D convolution | Use 3D convolution to process video data and complete behavior recognition |
It can be said that the difficulty is quite high. For images, the data is fixed, and what is captured is what it is! But text data is not so fixed, and sometimes it is not easy for humans to understand, let alone computers. High challenges are also high benefits. The development prospects of NLP are still very good. As for the specific direction to choose, it actually depends on everyone's preferences!
In 2018, a Google paper came out, BERT! It is equivalent to a general solution framework for natural language processing, and can basically do all tasks! This requires everyone to focus on learning, and it can be written as a project in your resume. It can be said to be one of the necessary skills for NLP today!
| Project name | Content overview |
|---|---|
| language model | The language model requires everyone to be familiar with the basis of subsequent word vectors. |
| Use Gemsim to build word vectors | Gensim is a really useful package! |
| Classification task based on word2vec | Let’s first use this example to understand how to use word vectors |
| Comparison of NLP-text feature methods | There are so many ways to construct text features, which one is better? |
| LSTM sentiment analysis | Use this project to understand what the input required by an RNN model looks like |
| NLP-similarity model | Text similarity calculation method |
| Conversational bot | Build a chatbot based on tensorlfow framework |
| Create your own input method | Can you build your own input method? Help you get it done! |
| Robot writes Tang poetry | Take a look at the Tang poetry written by the model! |
| NMT machine translation box | Open source project, capable of secondary development |
| Address Postcode Multiple Sequence Task | Classic text classification tasks |
| Principle of BERT, a general framework for natural language processing | This is the BERT mentioned above, the point! The point! The point! |
| Interpretation of the source code of Google's open source project BERT | Source code is very important, every line needs to be understood |
| Chinese emotional analysis based on BERT | Model development based on open source projects |
| Base -based Chinese naming physical identification | Named entity identification based on open source projects |
Through recent years of online courses and online business training, I have met many friends. The trust of institutions and students is that I will continue to update the biggest motivation of the course. Everyone knows that I basically go through video courses. I am very happy to bring everyone gains. I remember the most excited thing is to share with my family and have friends to gain offer. Thanks for the support of so many friends, come on, you are the best!
