morganwallace / classify_exercise Goto Github PK

#classify_exercise

This repository contains the scripts and documents for the UC Berkeley School of Information Spring 2014 Data Mining class' final project.

**Author: ** Morgan Wallace

Below is the proposal for this project

Morgan Wallace

***Please Note: ***

This project supports the work I am doing for my MIMS final project. The project is called *** Hercubit *** and it is a wearable fitness device that gives users real-time feedback on their computer to help them exercise.

To implement and begin to optimize a SVM classifier for free-weight exercises.

##Roles I, Morgan Wallace, am not working with any other student in Info 290T-03. I have collaborators on my MIMS final project team that I work closely with, however, the data mining related work that I submit for this class will be entirely my own. The entirety of my data will come from this other group project but I have been responsible for creating, cleaning, and organizing the data since the beginning, many months ago.

##Resources #####Data

My data is composed of 784 repititions of various free-weight exercises like:

• Bicep Curls
• Tricep Curls/Kickbacks
• Shoulder Press

Each repitition is made up of samples which are taken every 0.1 seconds and have raw values x, y, and z axes for each of the 3 sensors on the device (accelerometer, gyroscope, magnetometer).

The data is in the data/ directory and subdivided into

• labeled_data.csv - all data
• training_data.csv - all_data except holdout
• holdout.csv - for testing model on new data.

For a much more in-depth look at my data so far, please view one of my iPython notebooks:

• notebooks/SVC.ipynb - SVC classifier build and testing - also online here: http://nbviewer.ipython.org/github/morganwallace/classify_exercise/blob/master/notebooks/SVC.ipynb
• notebooks/Machine Learning.ipynb - Initial data prep and exploration - also online here: http://nbviewer.ipython.org/github/morganwallace/classify_exercise/blob/master/notebooks/Machine%20Learning%20-%20Free%20Weights.ipynb

rep_tracker_svc.py implements the SVC for real-time data coming from a Hercubit device (data/backup.csv will be iterated through instead of connecting to the device).

save_graph_and_data.py is the script that connects to the Hercubit device and visualizes and then saves the data for the training set. It will output the data/new_training.csv file that notebooks/SVC.ipynb uses to add new sets to the training dataset (although this output is disabled for this submission since you will not have a device - data/backup.csv will be iterated through instead).

hercubit/ directory is a package of scrips used for connected to a hercubit device. *For this submission, bluetooth has been disabled; data/backup.csv will be iterated through instead.

#####Software GitHub, Python, iPython and the following Python libraries:

• sklearn
• numpy
• matplotlib
• mpld3
• seaborn
• pandas
• numpy
• pyserial

#####Other I will be using the Hercubit wearable device as a source of data for the test, training and holdout datasets.