You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In this work, we present a new deep learning architecture for addressing the problem of supervised learning with sparse and irregularly sampled multivariate time
3
-
series. The architecture is based on the use of a semi-parametric interpolation
4
-
network followed by the application of a prediction network. The interpolation
5
-
network allows for information to be shared across multiple dimensions of a multivariate time series during the interpolation stage, while any standard deep learning model can be used for the prediction network.
6
-
7
-
We use a two layer interpolation network. The first interpolation layer performs a semi-parametric univariate interpolation for each of the D time series separately while the second layer merges information from across all of the D time series at each reference time point by taking into account the correlations among the time series.
8
-
9
-
10
-
## Reference
11
-
> Satya Narayan Shukla and Benjamin Marlin. Interpolation-prediction networks for irregularly sampled time series. In International Conference on Learning Representations, 2019. \[[pdf](https://openreview.net/pdf?id=r1efr3C9Ym)\]
12
-
2
+
This repository is an official implementation of [Interpolation-Prediction Networks for Irregularly Sampled Time Series](https://openreview.net/pdf?id=r1efr3C9Ym), accepted at ICLR 2019. In this work, we present a new deep learning architecture for addressing the problem of supervised learning with sparse and irregularly sampled multivariate time series.
3
+
<palign="center">
4
+
<imgwidth="600"src="imgs/model.png">
5
+
</p>
13
6
14
7
## Requirements
15
8
The code requires Python 3.7 or later. The file [requirements.txt](requirements.txt) contains the full list of
16
9
required Python modules.
10
+
```bash
11
+
pip install -r requirements.txt
12
+
```
17
13
18
-
## Usage
14
+
## Training and Evaluation
19
15
For running our model on univariate time series (UWave dataset):
The notations here align with the notation section 3.1 in the paper. For brevity, lets assume we have just one example in the training set and dimension `d = 2`. 
28
+
29
+
## Results
30
+
Classification and regression performance on MIMIC-III.
31
+
<palign="center">
32
+
<imgwidth="600"src="imgs/results.png">
33
+
</p>
34
+
35
+
## Data Format Example
36
+
The notations here align with the notation section 3.1 in the paper. For brevity, lets assume we have just one example in the training set and dimension `d = 2`.
37
+
<palign="center">
38
+
<imgwidth="600"src="imgs/dataformat.png">
39
+
</p>
40
+
41
+
## Reference
42
+
```bash
43
+
@inproceedings{
44
+
shukla2018interpolationprediction,
45
+
title={Interpolation-Prediction Networks for Irregularly Sampled Time Series},
46
+
author={Satya Narayan Shukla and Benjamin Marlin},
47
+
booktitle={International Conference on Learning Representations},
48
+
year={2019},
49
+
url={https://openreview.net/forum?id=r1efr3C9Ym},
50
+
}
51
+
```
34
52
35
53
## Contact
36
54
For more details, please contact <snshukla@cs.umass.edu>.
0 commit comments