Project Overview Welcome to the Convolutional Neural Networks (CNN) project in the AI Nanodegree! In this project, you will learn how to build a pipeline that can be used within a web or mobile app to process real-world, user-supplied images. Given an image of a dog, your algorithm will identify an estimate of the canine’s breed. If supplied an image of a human, the code will identify the resembling dog breed. Sample Output Along with exploring state-of-the-art CNN models for classification, you will make important design decisions about the user experience for your app. Our goal is that by completing this lab, you understand the challenges involved in piecing together a series of models designed to perform various tasks in a data processing pipeline. Each model has its strengths and weaknesses, and engineering a real-world application often involves solving many problems without a perfect answer. Your imperfect solution will nonetheless create a fun user experience! Project Instructions Instructions Clone the repository and navigate to the downloaded folder. git clone https://github.com/udacity/dog-project.git cd dog-project Download the dog dataset. Unzip the folder and place it in the repo, at location path/to/dog-project/dogImages. Download the human dataset. Unzip the folder and place it in the repo, at location path/to/dog-project/lfw. If you are using a Windows machine, you are encouraged to use 7zip to extract the folder. Download the VGG-16 bottleneck features for the dog dataset. Place it in the repo, at location path/to/dog-project/bottleneck_features. (Optional) If you plan to install TensorFlow with GPU support on your local machine, follow the guide to install the necessary NVIDIA software on your system. If you are using an EC2 GPU instance, you can skip this step. (Optional) If you are running the project on your local machine (and not using AWS), create (and activate) a new environment. Linux (to install with GPU support, change requirements/dog-linux.yml to requirements/dog-linux-gpu.yml): conda env create -f requirements/dog-linux.yml source activate dog-project Mac (to install with GPU support, change requirements/dog-mac.yml to requirements/dog-mac-gpu.yml): conda env create -f requirements/dog-mac.yml source activate dog-project NOTE: Some Mac users may need to install a different version of OpenCV conda install --channel https://conda.anaconda.org/menpo opencv3 Windows (to install with GPU support, change requirements/dog-windows.yml to requirements/dog-windows-gpu.yml): conda env create -f requirements/dog-windows.yml activate dog-project (Optional) If you are running the project on your local machine (and not using AWS) and Step 6 throws errors, try this alternative step to create your environment. Linux or Mac (to install with GPU support, change requirements/requirements.txt to requirements/requirements-gpu.txt): conda create --name dog-project python=3.5 source activate dog-project pip install -r requirements/requirements.txt NOTE: Some Mac users may need to install a different version of OpenCV conda install --channel https://conda.anaconda.org/menpo opencv3 Windows (to install with GPU support, change requirements/requirements.txt to requirements/requirements-gpu.txt): conda create --name dog-project python=3.5 activate dog-project pip install -r requirements/requirements.txt (Optional) If you are using AWS, install Tensorflow. sudo python3 -m pip install -r requirements/requirements-gpu.txt Switch Keras backend to TensorFlow. Linux or Mac: KERAS_BACKEND=tensorflow python -c "from keras import backend" Windows: set KERAS_BACKEND=tensorflow python -c "from keras import backend" (Optional) If you are running the project on your local machine (and not using AWS), create an IPython kernel for the dog-project environment. python -m ipykernel install --user --name dog-project --display-name "dog-project" Open the notebook. jupyter notebook dog_app.ipynb (Optional) If you are running the project on your local machine (and not using AWS), before running code, change the kernel to match the dog-project environment by using the drop-down menu (Kernel > Change kernel > dog-project). Then, follow the instructions in the notebook. NOTE: While some code has already been implemented to get you started, you will need to implement additional functionality to successfully answer all of the questions included in the notebook. Unless requested, do not modify code that has already been included. Evaluation Your project will be reviewed by a Udacity reviewer against the CNN project rubric. Review this rubric thoroughly, and self-evaluate your project before submission. All criteria found in the rubric must meet specifications for you to pass. Project Submission When you are ready to submit your project, collect the following files and compress them into a single archive for upload: The dog_app.ipynb file with fully functional code, all code cells executed and displaying output, and all questions answered. An HTML or PDF export of the project notebook with the name report.html or report.pdf. Any additional images used for the project that were not supplied to you for the project. Please do not include the project data sets in the dogImages/ or lfw/ folders. Likewise, please do not include the bottleneck_features/ folder.

Understanding gender differences in professional European football through Machine Learning interpretability and match actions data. This repository contains the full data pipeline implemented for the study *Understanding gender differences in professional European football through Machine Learning interpretability and match actions data*. We evaluated European male, and female football players' main differential features in-match actions data under the assumption of finding significant differences and established patterns between genders. A methodology for unbiased feature extraction and objective analysis is presented based on data integration and machine learning explainability algorithms. Female (1511) and male (2700) data points were collected from event data categorized by game period and player position. Each data point included the main tactical variables supported by research and industry to evaluate and classify football styles and performance. We set up a supervised classification pipeline to predict the gender of each player by looking at their actions in the game. The comparison methodology did not include any qualitative enrichment or subjective analysis to prevent biased data enhancement or gender-related processing. The pipeline had three representative binary classification models; A logic-based Decision Trees, a probabilistic Logistic Regression and a multilevel perceptron Neural Network. Each model tried to draw the differences between male and female data points, and we extracted the results using machine learning explainability methods to understand the underlying mechanics of the models implemented. A good model predicting accuracy was consistent across the different models deployed. ## Installation Install the required python packages ``` pip install -r requirements.txt ``` To handle heterogeneity and performance efficiently, we use PySpark from [Apache Spark](https://spark.apache.org/). PySpark enables an end-user API for Spark jobs. You might want to check how to set up a local or remote Spark cluster in [their documentation](https://spark.apache.org/docs/latest/api/python/index.html). ## Repository structure This repository is organized as follows: - Preprocessed data from the two different data streams is collecting in [the data folder](data/). For the Opta files, it contains the event-based metrics computed from each match of the 2017 Women's Championship and a single file calculating the event-based metrics from the 2016 Men's Championship published [here](https://figshare.com/collections/Soccer_match_event_dataset/4415000/5). Even though we cannot publish the original data source, the two python scripts implemented to homogenize and integrate both data streams into event-based metrics are included in [the data gathering folder](data_gathering/) folder contains the graphical images and media used for the report. - The [data cleaning folder](data_cleaning/) contains descriptor scripts for both data streams and [the final integration](data_cleaning/merger.py) - [Classification](classification/) contains all the Jupyter notebooks for each model present in the experiment as well as some persistent models for testing.

This repository contains python code and jupyter notebook for Image Classifier built using pytorch framework and VGG Deep Learning Image Classification Algorithm

ANALYZING ROAD SAFETY & TRAFFIC DEMOGRAPHICS IN THE UK (Multi-class Classification) SUMMARY Here, I am aim to analyze the Road Safety and Traffic Demographics dataset (UK), containing accidents reported by the police between the years of 2004 - 2017. PROJECT GOALS: Identify factors responsible for most of the reported accidents. Build a machine learning model that is capable of accurately predicting the severity of an accident. Provide recommendations to the Department of Transport (UK Government), to improve road safety policies and prevent recurrences of severe accidents where possible. PACKAGES USED: Scikit-learn, numpy, pandas, imblearn (imbalanced-learn), seaborn, Matplotlib MOTIVATION World Health Organization (WHO) reported that more than 1.25 million people die each year while 50 million are injured as a result of road accidents worldwide. Road accidents are the 10th leading cause of death globally. On current trends, road traffic accidents are to become the 7th leading cause of death by 2030 making it a major public health concern. Between the years 2005 and 2016, there were roughly 2 million road accidents reported in the United Kingdom (UK) alone of which 16,000 were fatal. As a big data project, I wanted to explore the traffic demographics data in greater detail using machine learning! CONTEXT The UK government amassed traffic data from 2004 to 2017, recording over 2 million accidents in the process and making this one of the most comprehensive traffic data sets out there. It's a huge picture of a country undergoing change. Note that all the contained accident data comes from police reports, so this data does not include minor incidents. For steps undertaken to pre-process and clean the data, please view the "Data Cleansing & Descriptive Analysis_UK Traffic Demographics.ipynb" file DESCRIPTIVE ANALYTICS (EDA) Tools used include Python, Tableau, MS PowerBI Percent (%) distribution of target classes Percent dist of Accident Severity As seen above, the data is highly imbalanced. For detailed steps undertaken to deal with the imbalanced data, please view the "Modelling_Predictive Analytics_UK Traffic Demographics.ipynb" file. This article provides some great tips on utilizing the correct performance metrics when analyzing a models performance trained on an imbalanced dataset. This article describes several strategies that can help combat the case of a severly imbalanced dataset. Methods include: Resampling strategies (under - Tomek Links, Cluster Centroids, over sampling - SMOTE) Using Decision Tree based models Using Cost-Sensitive training (Penalize algorithms) Number of accidents by Year and Accident Severity Total accidents by year and severity It can be seen above that the trend seems to be increasing as the years go. In addition, the spike between 2008 - 2009 was because of a enhancement in the reporting system introduced in the UK in 2009, where all accident including minor accidents needed to be reported by the police so as to match the counts represented by hospitals, insurance claims etc. Accidents density by Location geomap Most accidents took place in major cities - Birmingham, London, leeds, Newcastle Accidents by Gender and Age Accidents by gender and age Accidents by Day of the week and Year Accidents by year and weekday Most accidents take place on a Friday Vehicle Manoever at time of accident Vehicle Manoever at time of accident Most accidents take place as a result of overtaking For more findings, please go to the "Images" folder. For steps undertaken to carry out some predictive modeling and hyper-parameter tuning, please view the "Modelling_Predictive Analytics_UK Traffic Demographics.ipynb" file. RECOMMENDATIONS TO THE DEPARTMENT OF TRANSPORT (UK) Decrease emergency response times during afternoon rush-hours (15-19) especially on Fridays. Allocate resources to investigate high density traffic points and identify new infrastructure needs to divert traffic from dual-carriage ways. Explore conditions of vehicles and casualties such as vehicle type, age of vehicles registered, pedestrian movements, etc. for policy makers. Adopt comprehensive distracted driving laws that increase penalties for drivers who commit traffic violations like aggressive overtaking. ACKNOWLEDGEMENTS The license for this dataset is the Open Givernment Licence used by all data on data.gov.uk. The raw datasets are available from the UK Department of Transport website. I had a lot of fun working on this dataset and learned a lot in the process. I plan to further my research in the area of predictive modeling using imabalanced data and how to effectively build a highly robust model for future projects. About Here, I analyze the Road Safety and Traffic Demographics dataset (UK), containing accidents reported by the police between the years of 2004 - 2017. Topics accident-rate accident-severity imbalanced-data imbalanced-learning road-accident reported-accidents road-safety uk-government transport traffic-demographics severe-accidents pca classification Resources Readme Releases No releases published Packages No packages published Languages Jupyter Notebook 100.0% © 2020 GitHub, Inc.

The aim of my project falls under image classification as it attempts to match images. The algorithm takes image inputs from the user and then cycles through images stored in its database and try to predict probabilities that these faces match. This algorithm has a multitude of uses depending upon the need. For example, the algorithm can be implemented by law enforcement to match CCTV footage with potential perpetrators, either to check for past offenders or as proof during a trial. It could similarly be used by companies when hiring people to check if they have a criminal record. The inspiration for my project arose from my frustration with the increase in identity fraud and the lack of solutions for it. When companies hire people, their attempts to conduct a thorough background check can be manipulated very easily. To solve such issues, I came up with the idea to build a database which stores the image of past offenders which can be used to find criminal records of individuals.The algorithm was built using a jupyter notebook and python programming language. I used a variety of libraries like numpy, tensorflow and cv2 to construct the algorithm. The algorithm was built over a period of two months during which my mentor and I worked on the algorithm and discussed various options on methods of improvement. To train the algorithm I built a celebrity image bank because of the ease of finding pictures at different angles. In my training set, I’ve used five images of twenty different celebrities at different angles and time periods. The first step was to build the framework of the neural network, where I specified an optimization function, the number of hidden layers and a maximum iteration. The algorithm begins by storing the image information from the database in a pixel array for colored photographs. The image is then resized to suit the algorithm As the pixel info of the images is quite memory intensive, I decided to resize down to 250*250 pixels and standardize the pixel rgb values .The next step is to train the algorithm, which I have done by calibrating its input layer to accept image pixel vectors, and then providing the algorithm instructions for the multiple layers of processing i.e. learning or looking for repeated pixel patterns from the stored images, a probability vector is outputted. This probability vector calculates the ‘similarity’ of the input image that matches each celebrity in the database.The purpose of this algorithm was to control crime based on identity theft, identity fraud and help the law enforcement departments to capture criminals. As the output is a probability vector, even if they used a disguise etc, the algorithm’s prediction would not be extremely compromised.

Deliverable: This Jupyter notebook will help aspiring data scientists learn and practice the necessary python code needed for creating a world-class deep learning image classifier using a convolutional neural network deep learning algorithm.