{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"Introduction
\n\nWith the current state of machine learning technology, stable diffusion, and computer vision, people find new ways to satisfy their desire to contribute new knowledge surrounding the issue of AI’s potential. In particular, the processing of data has become one of the major sources of curiosity in the past decade (He et al., 2016). A community of aspiring data scientists and software developers known as “The Kaggle Community” has banded together to achieve newfound understanding of the current state of machine learning. Through the release of datasets, facilitation of competitions, and use of collaboration, Kaggle data scientists, or Kagglers, have made major strides in machine learning research.\n\nOne data format has been immensely popular among the Kaggle community. This data is known as image data. Image data is a visual form of data that has sparked interest among the Kaggle community and applied in countless ways. Specifically, image data requires fewer parameters to process, which can lead to more interpretive conclusions. Additionally, image data has little input that requires processing, which can make parsing slightly more efficient. Lastly, the trends that we find in one part of an image can help to infer conclusions about another image as the images are mostly interpretive but share patterns (Russakovsky et. al, 2016).\n\nSuch patterns allow one to determine trends that occur among image data and help to generalize trends.\nLiterature Review
\nOne popular example involving the use of image data in machine learning is that of the classic “Cats versus Dogs” example in data science (Lee, 2021). Within the Kaggle community, people have worked with images of cats and dogs to create elaborate visualizations, train powerful machine models, and propose revolutionary conclusions within the realm of data science.\n\nKagglers from all around the world participated collaboratively to explore the realms and applications of data science. Microsoft Research provided a cat and dog image data set that created a diverse challenge for Kagglers to solve (Microsoft Research, 2013). Numerous different solutions were proposed by Kagglers such as the neural networks. \nWith the temptation of ASPCA animal charity donations, the competition garnered the attention of more than two hundred teams. This demonstrated the passion that the Kaggle community had for the project they pursued.\n\nWhat needs to be determined, however, is how the Kaggle community’s interaction with image data has had significance on gaining new understandings and making new progress (Yang et al., 2018). As for the performance of the models, the main knowledge gained is the understanding of the data and how it affects us in real time. Through the experience of analyzing and visualizing data with Python, one gains valuable skills that can help to develop relevant research and analysis skills. The community has developed many new different techniques throughout the years with members gaining valuable skills for their future.\n\nMethodology
\nThrough Kaggle, many image/video datasets will be identified. User submissions will be analyzed on a couple of factors including the purpose, the analysis, and the conclusions. This will help to determine how users interact with Kaggle data. The data types will determine how the analysis will be classified into groups. Each will have different ways to analyze, and they will hence be used in different ways. Additionally, the training algorithms for each set of code with identical data will differ as well. As a result, there will be drastically different results generated from different competitions, data types, and machine learning models.\n","metadata":{}},{"cell_type":"markdown","source":"\n\nHence, to study the hypothesis further, image and video data sets from competition datasets on Kaggle will be examined. Through the data set diversity, more information can be acquired about the significance of the data, and validify the credibility of such data. Code samples will help to determine the diversity of user experience on the Kaggle platform, and it will help to form conclusions about the original intentions of using Kaggle. This meta-analysis will centralize around four pillars: purpose, method, findings, and experience.","metadata":{}},{"cell_type":"markdown","source":"The primary data examined will be that of the Dog Identification dataset (Cukierski, 2016), the Help Protect the Great Barrier Reef dataset (Yang, 2021), and the Bundesliga Data Shootout dataset (Detsche Fußball Liga e.V., 2022). Through the extensive studies and analyses performed by Kaggle users on diverse image/video data sources, there have been many different issues solved, and many different skills learned. \\\nMany would agree that the new information learned by the data we have today can provide future insight into the progress of technology and its potential (Agarwal & Dhar, 2014). Hence, this analysis will focus on aspects of the intentions of Kaggle users when analyzing the trends and applications of the data use and application. \\\nWithin the four data sets and their analyses, a successful meta-analysis would identify each of the four pillars of purpose, method, findings, and experience within each of the sets of data. This will help to answer the primary question of how Kagglers interact with image data, why they interact with image data, and what they learn when they analyze image data.\\\nTo create the most effective analysis, the analysis will be performed in order of when the competition occurred to see the evolution of Kaggle and what Kaggle users learned from the data.\n","metadata":{}},{"cell_type":"markdown","source":"Analysis
","metadata":{}},{"cell_type":"markdown","source":"There are a couple of notebooks that encompass both the quality of the findings and the technologies used. These notebooks were published to Kaggle competitions by numerous avid Kagglers. The three notebooks highlighted the power of data science to analyze the features of the data that displayed significance.","metadata":{}},{"cell_type":"markdown","source":"Dog Breed Identifier
","metadata":{}},{"cell_type":"markdown","source":"One of these notebooks that emphasized the significance of feature engineering (See Appendix A) for image data was that of the Fast.ai analysis to identify dog breeds by a user by the name of KKS14 . This method was especially creative because it displayed the ability of the AI to accomplish the desired task. The benefit of this solution was the ingenuity of the solution. Most of the other notebooks in the competition used solutions involving computer vision engineering or Tensorflow. This was a risky move, as Fast.ai has very little documentation, and it relies way too much on Pytorch, which isn’t sponsored by as many for-profit organizations as TensorFlow. However, given the limitations of his method, it was a well-produced algorithm that identified the dog breeds well given the immense number of simplifications that the process went through. \nThrough the process of fast.ai though, KKS14 was able to manipulate Convolutional Neural Networks (CNNs) to allow the computer to see. Given that there are over twenty-five thousand animal photos in the data set, it makes sense to use a neural network to avoid the magnification of bias. As a result, fast.ai was able to classify dog breeds at 80% accuracy as of 2013.\n","metadata":{}},{"cell_type":"markdown","source":"\n","metadata":{}},{"cell_type":"markdown","source":"With this amount of accuracy, it can be determined that for the resources used, it was a good algorithm to classify the information. Given the benefits and limitations of Fast.ai, it was a risky choice to build a model based on Fast.ai, but it highlighted the curiosity of Kagglers to seek unique solutions.","metadata":{}},{"cell_type":"markdown","source":"Great Barrier Reef COTS Detector
","metadata":{}},{"cell_type":"markdown","source":"Another Kaggle competition that had some interesting solutions was the TensorFlow protect the Great Barrier Reef competition. With one-hundred-fifty thousand dollars in the prize pool, this competition attracted over two thousand unique submissions and multiple different solutions. This competition was a way to demonstrate computer vision skills. In this competition, users implemented a model to detect invasive blooms of Crown-of-thorns Starfish (COTS). The COTS blooms were a primary cause for reef habitat destruction, especially in areas like the Great Barrier Reef of Australia. The reef destruction led to Tensorflow to release a Kaggle competition to help develop a detection algorithm for the reef. Through this competition, many good algorithms were generated that allowed for the creation of a consistently accurate detection method that could detect nearly all of the stars.\\\nThe most notable notebooks within this competition came from the user KS14 who had a creative way of visualizing the data. Given that the only data given were videos and images extracted from the videos, one had to get creative to extrapolate the data from a qualitative image into a quantitative chart.\\\nThat required the use of OpenCV computer vision library for boxing which led to the generation of bboxes. Once bboxes were generated, the COTS could be easily detected. While YOLO deep learning algorithms and TensorFlow models were the most popular, the choice to use OpenCV led to an unconventional solution to the problem. Additionally, the user was able to graph the timestamps in a way that made it easy to understand (See Appendix B). Hence, it led to more learning opportunities for machine learning. This goes to show that the whole goal of competitions is less about the prize and more about learning collaboratively. \n","metadata":{}},{"cell_type":"markdown","source":"","metadata":{}},{"cell_type":"markdown","source":"Bundesliga Data Shootout
\nOne last competition that stood out was the Bundesliga Data Shootout. Sponsored by the Bundesliga, the competition was hosted to quickly use machine learning to classify different plays in football like passes, throw-ins, and crosses. \nThis model was especially difficult to implement because one would not only have to identify where players are, but where the ball is and where the goals are. Hence, the scores were not as high as other competitions.\n\nHowever, one solution by user RM1000 showcased what it meant to visualize a problem really well. For this notebook, its purpose wasn’t to create a model, but rather understanding the data and the problem first before proposing a solution. As a result, one can see what is documented for each video frame.\n\nUsing the classic Python libraries of matplotlib and seaborn, one can generate great visualizations that can better help with seeing data before one attempt to create a model using it. As a result, data visualization is helpful for those who want to understand the data, but also learn more about the data too.\n\nWith data visualization, it makes it easier for people to understand the solution as well as the data. Since the Kaggle community has always been about learning and bettering as a community, this notebook was especially good for that task.\n\nWhile the higher scoring notebooks tended to use more conventional models like TensorFlow or YOLO, RM1000 used an OpenCV detection algorithm, which would allow for quick real-time AI vision. Hence, it would make it faster, but it wouldn’t guarantee a higher accuracy score. Taking this move was risky, but it showed that creative solutions are always welcome in the Kaggle community, and they are highly encouraged.\n","metadata":{}},{"cell_type":"markdown","source":"Conclusion
\nIn retrospect, the way that the Kaggle community has interacted with image data has shown that its capability to come together as a whole to work toward a common goal is strong. With three different data imaging projects, it is especially apparent that the Kaggle community cares a lot about collaborative learning and sharing understanding. Especially with image data, there were numerous different computer vision solutions that made an impact on everyday life today.\n\nThe way that the Kaggle community can find multiple different solutions to solve problems that are essentially the same is extraordinary.\n","metadata":{}},{"cell_type":"markdown","source":"Appendix A
","metadata":{}},{"cell_type":"markdown","source":"","metadata":{}},{"cell_type":"markdown","source":"Appendix B
","metadata":{}},{"cell_type":"markdown","source":"","metadata":{}},{"cell_type":"markdown","source":"Appendix C
","metadata":{}},{"cell_type":"markdown","source":"","metadata":{}},{"cell_type":"markdown","source":"Author's Note: PII like names have been Pseudonymized for this competition to protect privacy. If you would like credit for your work verbally, please message me.","metadata":{}},{"cell_type":"markdown","source":"References
\nAgarwal, R., & Dhar, V. (2014). Big data, data science, and analytics: The opportunity and challenge for IS Research. Information Systems Research, 25(3), 443–448. https://doi.org/10.1287/isre.2014.0546 \n\n\nChoetkiertikul, M., Hoonlor, A., Ragkhitwetsagul, C., Pongpaichet, S., Sunetnanta, T., Settewong, T., Jiravatvanich, V., & Kaewpichai, U. (2023, April 11). Mining the characteristics of Jupyter notebooks in Data Science Projects. arXiv.org. https://arxiv.org/abs/2304.05325 \n\nCukierski. (2016). Dog breed identification using deep learning. https://www.researchgate.net/publication/328834665_Dog_Breed_Identification_Using_Deep_Learning \n\nDetsche Fußball Liga e.V. (2022). DFL - Bundesliga Data Shootout. Deutsche Fussball Liga. https://www.kaggle.com/competitions/dfl-bundesliga-data-shootout/overview \n\nHe, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 770–778. https://doi.org/10.1109/cvpr.2016.90 \n\nLee, Y. (2021). Image classification with Artificial Intelligence: Cats Vs Dogs. 2021 2nd International Conference on Computing and Data Science (CDS). https://doi.org/10.1109/cds52072.2021.00081 \n\nMicrosoft Research. (2013). Cats and Dogs Dataset. Kaggle Cats and Dogs Dataset from Official Microsoft Download Center. https://www.microsoft.com/en-us/download/confirmation.aspx?id=54765 \n\nRussakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., Berg, A. C., & Fei-Fei, L. (2015). Imagenet Large Scale Visual Recognition Challenge. International Journal of Computer Vision, 115(3), 211–252. https://doi.org/10.1007/s11263-015-0816-y \n\nYang, L., Zhang, Y., Chen, H., Gao, Q., Luo, P., Hu, J., Zheng, F., Yuan, Y., Fu, L., & Chen, C. (2022). Physiological and transcriptomic responses to starvation in the corallivorous crown-of-thorn starfish. Frontiers in Marine Science, 9. https://doi.org/10.3389/fmars.2022.1021377 \n\nYang, X., Zeng, Z., Teo, S. G., Wang, L., Chandrasekhar, V., & Hoi, S. (2018). Deep learning for practical image recognition. Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 923–931. https://doi.org/10.1145/3219819.3219907 \n\n","metadata":{}},{"cell_type":"code","source":"import pandas as pd\nfilename = \"/kaggle/input/2023-kaggle-ai-report/sample_submission.csv\"\n\nsubmission = pd.read_csv(filename)\nsubmission.head()","metadata":{"execution":{"iopub.status.busy":"2023-07-04T01:10:34.743962Z","iopub.execute_input":"2023-07-04T01:10:34.744714Z","iopub.status.idle":"2023-07-04T01:10:34.813990Z","shell.execute_reply.started":"2023-07-04T01:10:34.744674Z","shell.execute_reply":"2023-07-04T01:10:34.812823Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission.loc[0]['value'] = 'Image and/or video data'\nsubmission.loc[1]['value'] = 'https://www.kaggle.com/code/ethanhu8093/meta-analysis-on-kaggle-computer-vision-approc/input?select=sample_submission.csv'\nsubmission.loc[2]['value'] = 'https://www.kaggle.com/code/thedrcat/how-to-win-a-kaggle-competition/comments#2328902'\nsubmission.loc[3]['value'] = 'https://www.kaggle.com/code/christph/copyright-and-privacy-issues-of-generative-ai/comments#2328913'\nsubmission.loc[4]['value'] = 'https://www.kaggle.com/code/sriramm2010/autonomous-cars-a-pinnacle-in-computer-vision/comments#2328918'\n\nsubmission.head()","metadata":{"execution":{"iopub.status.busy":"2023-07-04T01:23:32.454735Z","iopub.execute_input":"2023-07-04T01:23:32.455151Z","iopub.status.idle":"2023-07-04T01:23:32.468532Z","shell.execute_reply.started":"2023-07-04T01:23:32.455121Z","shell.execute_reply":"2023-07-04T01:23:32.467263Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"submission.to_csv('submission.csv',index=False)","metadata":{"execution":{"iopub.status.busy":"2023-07-04T01:24:02.278864Z","iopub.execute_input":"2023-07-04T01:24:02.279285Z","iopub.status.idle":"2023-07-04T01:24:02.286039Z","shell.execute_reply.started":"2023-07-04T01:24:02.279249Z","shell.execute_reply":"2023-07-04T01:24:02.284810Z"},"trusted":true},"execution_count":null,"outputs":[]}]}