Datasets and Software
Some of our datasets, software packages, and the code we use to run
our experiments may be of interest to other SENSOPAC partners. During
the course of the project, we make relevant material available here.
Learning Kinematics and Dynamics
Click here to view an introductory page to this
subproject.
Software package: Locally Weighted Projection Regression (LWPR)
We have implemented the LWPR algorithm as an ANSI-C library, with
convenient wrappers for Matlab and Python. The LWPR software package
has its own webpage, which you can access
here.
Randomly Varying Coefficient model
RVC uses a Bayesian formulation of local learning, which yields a
principled approach to dealing with open parameters. We've made available
a current version (0.1) of our implementations in Matlab and C++.
The archives also includes a toy dataset and a litte demo.
Notice: The C++ version makes use of the BOOST library, which you'll have
to download and install yourself.
Context Estimation and Switching
Click here to view an introductory page to this
subproject.
Multiple discrete models - additional material for Deliverable D2.2
The following datasets stem from experiments using a simulated 3-DoF robot arm
with a discrete set of attached loads. We split the data into numerous ZIP archives,
which should be extracted into the same directory.
File | Description |
Scripts.zip |
Description of the file formats and useful MATLAB scripts |
LearnStationary_Scripts.zip |
Scripts used for learning multiple inverse dynamics models |
LearnStationary_*.zip
1,
2,
3,
4,
5,
6,
7,
8,
9,
10 |
Training data for contexts 1-10 Each file is about 13 MByte |
ConEstDiscrete_Scripts.zip |
Scripts used for estimating the context from a set of learned models |
ConEstDiscrete_NoControl_NoHMM_NoConf_*.zip
Run 1,
Run 2,
Run 3,
Run 4,
Run 5 |
Context estimation without using control, HMM modelling and LWPR confidence bounds. Each file is about 20 MByte |
ConEstDiscrete_NoControl_NoHMM_Conf_*.zip
Run 1,
Run 2,
Run 3,
Run 4,
Run 5 |
Context estimation without using control and HMM modelling, but using LWPR confidence bounds. Each file is about 20 MByte |
ConEstDiscrete_NoControl_HMM_NoConf_*.zip
Run 1,
Run 2,
Run 3,
Run 4,
Run 5 |
Context estimation without using control and LWPR confidence bounds, but using HMM modelling. Each file is about 20 MByte |
ConEstDiscrete_NoControl_HMM_Conf_*.zip
Run 1,
Run 2,
Run 3,
Run 4,
Run 5 |
Context estimation without using control, but using HMM modelling and LWPR confidence bounds. Each file is about 20 MByte |
ConEstDiscrete_Control_NoHMM_NoConf_*.zip
Run 1,
Run 2,
Run 3,
Run 4,
Run 5 |
Context estimation using control, no HMM modelling, no LWPR confidence bounds. Each file is about 20 MByte |
ConEstDiscrete_Control_NoHMM_Conf_*.zip
Run 1,
Run 2,
Run 3,
Run 4,
Run 5 |
Context estimation using control, no HMM modelling, but using LWPR confidence bounds. Each file is about 20 MByte |
ConEstDiscrete_Control_HMM_NoConf_*.zip
Run 1,
Run 2,
Run 3,
Run 4,
Run 5 |
Context estimation using control, HMM modelling, but no LWPR confidence bounds. Each file is about 20 MByte |
ConEstDiscrete_Control_HMM_Conf_*.zip
Run 1,
Run 2,
Run 3,
Run 4,
Run 5 |
Context estimation using control, HMM modelling and LWPR confidence bounds. Each file is about 20 MByte |
|