Safely - The Intelligent Travel Assistant

2 minute read

Published: October 13, 2019

Our submission on Devpost and ML report

Inspiration

Nowadays people travel without awareness. We would like to make travels safer, using Big Data and AI.

What it does

Our model predicts the severity of accidents. We also created a web app that suggests useful tips to make your travel safer.

How we built it

We used many advance ML techniques and frameworks to process data and perform classification using Python language and the libraries written in it. We built an app using modern and fast searching engines as Algolia and Open Street Maps API. We have also implemented a simple and accessible UI design with Angular and Google Maps API.

Challenges we ran into.

We have a lot of problems with the dimension reduction of the dataset.
Setting a parameter space for GridSearch was also a big challenge.
Feature processing and training were very time-consuming.
We had to solve problem with Open Street Routing Machine - the demo which allows finding the route between two points has limits, so we have deployed the webservice on the AWS virtual machine.
The use of APIs was difficult because some of them have documentation which is hard to understand and our goal was tougher to achieve.
The way to display complex results in an accessible way for everyone was also a brainstorming challenge.

Accomplishments that we’re proud of

Creating an ML application capable to achieve such a good score in the classification task.
Creating a Web App that uses the predictions to help people to travel safer.
The method to transform from points to routes and the way we implemented the accident geo search.
We created a cool group of friends and meet a lot of interesting people during HackUPC 2019.

What we learned

Try to use PCA and other dimension reducing techniques as LDA or QDA instead of feature selection.
Know your algorithm parameters to properly GridSearch over them.
Feature processing may be very time-consuming use functions and generalize your tasks.
Consuming API using Angular and interact with different APIs at the same time.
Simple and useful UI project designing.
Sleep at least 4 hours a day to be more productive.
How to configure CodeBuild and CodePipeline in AWS Platform.

What’s next for Safely

We will expand it to other countries, we will also add other sources of data and the newest technology to make your travel even safer.

Built With

Check out GitHub Repo and HackUPC 2019

HackUPC-279

HackUPC-1210

HackUPC-1229

Photos Team Hack UPC

13 Oct 2019 Mateusz Dorobek

Share on

Twitter Facebook LinkedIn

Statystyka - a komu to potrzebne? A dlaczego?

4 minute read

Published: July 09, 2020

W tym poście chciałbym przybliżyć temat statystyki dla tych osób które nie miały z nią do czynienia, ale także tym którzy posługują się pewnymi narzędziami ale niewiedzą jaka intuicja za nimi stoi. Ten post będzie być może początkiem dłuższej serii o tym jak ja rozumiem tę dziedzinę matematyki. Read more

Master Thesis: Generating Jazz Chords Progressions Using Word Embeddings and Recurrent Neural Networks

1 minute read

Published: June 19, 2020

This thesis describes the problem of generating jazz music chords using recurrent neural networks (RNN) and numerical vector representations - embeddings. To create the embedding I use techniques known from the field of natural language processing (NLP). It is possible because of certain structural similarities between spoken language and music. I study the relationship between the chords in the hidden space generated by the algorithms: Word2Vec, FastText and Multi-hot. Visualization of chords in reduced vector representation space using t-SNE and PCA algorithms illustrates a lot of dependencies between the chords derived from the principles of jazz harmony. The test results confirmed the highest performance of chord generation when using the Word2Vec model in the Skip-Gram variant. To generate chords I use their vector representation. I analyze the performance of sixteen models based on the recursive neural networks in terms of their hyper parameters and architecture. I also study the impact of the algorithm used to create the numerical representation on the results of recurrent models. To train neural networks I use a set of jazz standards obtained from publicly available Internet sources in a format allowing for the extraction of chord sequences. The experiments shows the highest performance of deep models with recurrent layers - GRU and LSTM, but the one with the shortest runtime is the model with two GRU layers, Dropout regularization layer and Dense layer. I use trained models to build a simple program capable of generating continuous chord sequences in the unsupervised variant and with partially controlling this generation by the user. For implementation of all algorithms, visualization and data processing I use Python language and its libraries: tensorflow, keras, gensim, scikit-learn, numpy, pychord, music21 etc. Read more

Mateusz Dorobek