Datasets:
Formats:
parquet
Size:
100M - 1B
ArXiv:
Tags:
timeseries
timeseries clustering
changepoint-detection
correlation-structure
Synthetic
benchmark
License:
Updated description and added yaml data for the data viewer
Browse files
README.md
CHANGED
@@ -1,48 +1,473 @@
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
4 |
# CSTS - Correlation Structures in Time Series
|
5 |
|
6 |
## Important Notice
|
7 |
-
This dataset is published as a pre-publication release. An accompanying research paper is forthcoming on arXiv.
|
|
|
8 |
|
9 |
## Dataset Description
|
10 |
CSTS (**C**orrelation **S**tructures in **T**ime **S**eries) is a comprehensive synthetic benchmarking dataset for
|
11 |
evaluating correlation structure discovery in time series data. The dataset systematically models known correlation
|
12 |
structures between three different time series variates and enables examination of how these structures are affected by
|
13 |
distribution shifting, sparsification, and downsampling. With its controlled properties and ground truth labels,
|
14 |
-
CSTS provides algorithm developers clean benchmark data that
|
15 |
and messy real-world data.
|
16 |
|
17 |
### Key Applications
|
18 |
-
- Evaluating the ability of time series clustering algorithms to segment and group by correlation structures
|
19 |
-
- Assessing clustering validation methods for correlation-based clusters
|
20 |
-
- Investigating how data preprocessing affects correlation structure discovery
|
21 |
-
- Establishing performance thresholds for high-quality clustering result
|
22 |
|
23 |
### Dataset Structure
|
24 |
CSTS provides **two main splits** (exploratory and confirmatory) with **30 subjects** each, enabling proper statistical validation.
|
25 |
The dataset structure includes:
|
26 |
-
- **12 data variants
|
27 |
-
- **Generation stages**: raw (unstructured), correlated (normal-distributed),
|
28 |
-
- **Completeness levels**: complete (100%), partial (70%), sparse (10%
|
29 |
|
30 |
### Subjects
|
31 |
-
Each subject contains 100 segments of varying lengths and each segment encodes one of the 23 specific correlation
|
32 |
-
structures. Each subject uses all 23 patterns 4-5 times.
|
|
|
33 |
|
34 |
-
|
35 |
-
- a time series **data file** with three variates
|
36 |
- a **label file** specifying the ground truth segmentation and clustering
|
37 |
-
- 67 **bad clustering label files** with controlled degradations (varying numbers of segmentations and/or cluster assignment mistakes) spanning the
|
38 |
|
39 |
### Additional Splits
|
40 |
-
CSTS also includes versions (configured as splits) that allow
|
41 |
-
algorithm performance. They follow the same structure
|
42 |
-
|
43 |
-
- Reduced
|
|
|
44 |
|
45 |
-
Our accompanying paper provides complete methodological details, baseline findings, and usage guidance.
|
|
|
|
|
|
|
|
|
46 |
|
47 |
## Authors
|
48 |
- Isabella Degen, University of Bristol
|
@@ -50,7 +475,6 @@ Our accompanying paper provides complete methodological details, baseline findin
|
|
50 |
- Henry W J Reeve, University of Nanjing
|
51 |
- Kate Robson Brown, University College Dublin
|
52 |
|
53 |
-
|
54 |
## Pre-Publication Release Details
|
55 |
- **Release Date:** 29 Apr 2024
|
56 |
- **Version:** 1.0-pre
|
@@ -79,5 +503,5 @@ Please use the following temporary citation until our paper is published:
|
|
79 |
Once our paper is published on arXiv, we will update this README with the proper citation information.
|
80 |
**Please check back for updates.**
|
81 |
|
82 |
-
##
|
83 |
-
...coming soon
|
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
3 |
+
configs:
|
4 |
+
- config_name: raw_complete_data
|
5 |
+
data_files:
|
6 |
+
- split: exploratory
|
7 |
+
path: "exploratory/raw/*-data.parquet"
|
8 |
+
features: &id001
|
9 |
+
- name: subject_id
|
10 |
+
dtype: string
|
11 |
+
- name: datetime
|
12 |
+
dtype: string
|
13 |
+
type: timestamp
|
14 |
+
- name: iob
|
15 |
+
dtype: float32
|
16 |
+
- name: cob
|
17 |
+
dtype: float32
|
18 |
+
- name: ig
|
19 |
+
dtype: float32
|
20 |
+
- config_name: raw_complete_labels
|
21 |
+
data_files:
|
22 |
+
- split: exploratory
|
23 |
+
path: "exploratory/raw/*-labels.parquet"
|
24 |
+
features: &id002
|
25 |
+
- name: subject_id
|
26 |
+
dtype: string
|
27 |
+
- name: id
|
28 |
+
dtype: int32
|
29 |
+
- name: start idx
|
30 |
+
dtype: int32
|
31 |
+
- name: end idx
|
32 |
+
dtype: int32
|
33 |
+
- name: length
|
34 |
+
dtype: int32
|
35 |
+
- name: cluster_id
|
36 |
+
dtype: int32
|
37 |
+
- name: correlation to model
|
38 |
+
dtype: string
|
39 |
+
- name: correlation achieved
|
40 |
+
dtype: string
|
41 |
+
- name: correlation achieved with tolerance
|
42 |
+
dtype: string
|
43 |
+
- name: MAE
|
44 |
+
dtype: float32
|
45 |
+
- name: relaxed MAE
|
46 |
+
dtype: float32
|
47 |
+
- config_name: raw_complete_badclustering_labels
|
48 |
+
data_files:
|
49 |
+
- split: exploratory
|
50 |
+
path: "exploratory/raw/bad_partitions/*-labels.parquet"
|
51 |
+
features: &id003
|
52 |
+
- name: subject_id
|
53 |
+
dtype: string
|
54 |
+
- name: cluster_desc
|
55 |
+
dtype: string
|
56 |
+
- name: id
|
57 |
+
dtype: int32
|
58 |
+
- name: start idx
|
59 |
+
dtype: int32
|
60 |
+
- name: end idx
|
61 |
+
dtype: int32
|
62 |
+
- name: length
|
63 |
+
dtype: int32
|
64 |
+
- name: cluster_id
|
65 |
+
dtype: int32
|
66 |
+
- name: correlation to model
|
67 |
+
dtype: string
|
68 |
+
- name: correlation achieved
|
69 |
+
dtype: string
|
70 |
+
- name: correlation achieved with tolerance
|
71 |
+
dtype: string
|
72 |
+
- name: MAE
|
73 |
+
dtype: float32
|
74 |
+
- name: relaxed MAE
|
75 |
+
dtype: float32
|
76 |
+
- config_name: raw_partial_data
|
77 |
+
data_files:
|
78 |
+
- split: exploratory
|
79 |
+
path: "exploratory/irregular_p30/raw/*-data.parquet"
|
80 |
+
features: *id001
|
81 |
+
- config_name: raw_partial_labels
|
82 |
+
data_files:
|
83 |
+
- split: exploratory
|
84 |
+
path: "exploratory/irregular_p30/raw/*-labels.parquet"
|
85 |
+
features: *id002
|
86 |
+
- config_name: raw_partial_badclustering_labels
|
87 |
+
data_files:
|
88 |
+
- split: exploratory
|
89 |
+
path: "exploratory/irregular_p30/raw/bad_partitions/*-labels.parquet"
|
90 |
+
features: *id003
|
91 |
+
- config_name: raw_sparse_data
|
92 |
+
data_files:
|
93 |
+
- split: exploratory
|
94 |
+
path: "exploratory/irregular_p90/raw/*-data.parquet"
|
95 |
+
features: *id001
|
96 |
+
- config_name: raw_sparse_labels
|
97 |
+
data_files:
|
98 |
+
- split: exploratory
|
99 |
+
path: "exploratory/irregular_p90/raw/*-labels.parquet"
|
100 |
+
features: *id002
|
101 |
+
- config_name: raw_sparse_badclustering_labels
|
102 |
+
data_files:
|
103 |
+
- split: exploratory
|
104 |
+
path: "exploratory/irregular_p90/raw/bad_partitions/*-labels.parquet"
|
105 |
+
features: *id003
|
106 |
+
- config_name: correlated_complete_data
|
107 |
+
data_files:
|
108 |
+
- split: exploratory
|
109 |
+
path: "exploratory/normal/*-data.parquet"
|
110 |
+
features: *id001
|
111 |
+
- config_name: correlated_complete_labels
|
112 |
+
data_files:
|
113 |
+
- split: exploratory
|
114 |
+
path: "exploratory/normal/*-labels.parquet"
|
115 |
+
features: *id002
|
116 |
+
- config_name: correlated_complete_badclustering_labels
|
117 |
+
data_files:
|
118 |
+
- split: exploratory
|
119 |
+
path: "exploratory/normal/bad_partitions/*-labels.parquet"
|
120 |
+
features: *id003
|
121 |
+
- config_name: correlated_partial_data
|
122 |
+
data_files:
|
123 |
+
- split: exploratory
|
124 |
+
path: "exploratory/irregular_p30/normal/*-data.parquet"
|
125 |
+
features: *id001
|
126 |
+
- config_name: correlated_partial_labels
|
127 |
+
data_files:
|
128 |
+
- split: exploratory
|
129 |
+
path: "exploratory/irregular_p30/normal/*-labels.parquet"
|
130 |
+
features: *id002
|
131 |
+
- config_name: correlated_partial_badclustering_labels
|
132 |
+
data_files:
|
133 |
+
- split: exploratory
|
134 |
+
path: "exploratory/irregular_p30/normal/bad_partitions/*-labels.parquet"
|
135 |
+
features: *id003
|
136 |
+
- config_name: correlated_sparse_data
|
137 |
+
data_files:
|
138 |
+
- split: exploratory
|
139 |
+
path: "exploratory/irregular_p90/normal/*-data.parquet"
|
140 |
+
features: *id001
|
141 |
+
- config_name: correlated_sparse_labels
|
142 |
+
data_files:
|
143 |
+
- split: exploratory
|
144 |
+
path: "exploratory/irregular_p90/normal/*-labels.parquet"
|
145 |
+
features: *id002
|
146 |
+
- config_name: correlated_sparse_badclustering_labels
|
147 |
+
data_files:
|
148 |
+
- split: exploratory
|
149 |
+
path: "exploratory/irregular_p90/normal/bad_partitions/*-labels.parquet"
|
150 |
+
features: *id003
|
151 |
+
- config_name: nonnormal_complete_data
|
152 |
+
data_files:
|
153 |
+
- split: exploratory
|
154 |
+
path: "exploratory/non_normal/*-data.parquet"
|
155 |
+
features: *id001
|
156 |
+
- config_name: nonnormal_complete_labels
|
157 |
+
data_files:
|
158 |
+
- split: exploratory
|
159 |
+
path: "exploratory/non_normal/*-labels.parquet"
|
160 |
+
features: *id002
|
161 |
+
- config_name: nonnormal_complete_badclustering_labels
|
162 |
+
data_files:
|
163 |
+
- split: exploratory
|
164 |
+
path: "exploratory/non_normal/bad_partitions/*-labels.parquet"
|
165 |
+
features: *id003
|
166 |
+
- config_name: nonnormal_partial_data
|
167 |
+
data_files:
|
168 |
+
- split: exploratory
|
169 |
+
path: "exploratory/irregular_p30/non_normal/*-data.parquet"
|
170 |
+
features: *id001
|
171 |
+
- config_name: nonnormal_partial_labels
|
172 |
+
data_files:
|
173 |
+
- split: exploratory
|
174 |
+
path: "exploratory/irregular_p30/non_normal/*-labels.parquet"
|
175 |
+
features: *id002
|
176 |
+
- config_name: nonnormal_partial_badclustering_labels
|
177 |
+
data_files:
|
178 |
+
- split: exploratory
|
179 |
+
path: "exploratory/irregular_p30/non_normal/bad_partitions/*-labels.parquet"
|
180 |
+
features: *id003
|
181 |
+
- config_name: nonnormal_sparse_data
|
182 |
+
data_files:
|
183 |
+
- split: exploratory
|
184 |
+
path: "exploratory/irregular_p90/non_normal/*-data.parquet"
|
185 |
+
features: *id001
|
186 |
+
- config_name: nonnormal_sparse_labels
|
187 |
+
data_files:
|
188 |
+
- split: exploratory
|
189 |
+
path: "exploratory/irregular_p90/non_normal/*-labels.parquet"
|
190 |
+
features: *id002
|
191 |
+
- config_name: nonnormal_sparse_badclustering_labels
|
192 |
+
data_files:
|
193 |
+
- split: exploratory
|
194 |
+
path: "exploratory/irregular_p90/non_normal/bad_partitions/*-labels.parquet"
|
195 |
+
features: *id003
|
196 |
+
- config_name: downsampled_complete_data
|
197 |
+
data_files:
|
198 |
+
- split: exploratory
|
199 |
+
path: "exploratory/resampled_1min/*-data.parquet"
|
200 |
+
features: *id001
|
201 |
+
- config_name: downsampled_complete_labels
|
202 |
+
data_files:
|
203 |
+
- split: exploratory
|
204 |
+
path: "exploratory/resampled_1min/*-labels.parquet"
|
205 |
+
features: *id002
|
206 |
+
- config_name: downsampled_complete_badclustering_labels
|
207 |
+
data_files:
|
208 |
+
- split: exploratory
|
209 |
+
path: "exploratory/resampled_1min/bad_partitions/*-labels.parquet"
|
210 |
+
features: *id003
|
211 |
+
- config_name: downsampled_partial_data
|
212 |
+
data_files:
|
213 |
+
- split: exploratory
|
214 |
+
path: "exploratory/irregular_p30/resampled_1min/*-data.parquet"
|
215 |
+
features: *id001
|
216 |
+
- config_name: downsampled_partial_labels
|
217 |
+
data_files:
|
218 |
+
- split: exploratory
|
219 |
+
path: "exploratory/irregular_p30/resampled_1min/*-labels.parquet"
|
220 |
+
features: *id002
|
221 |
+
- config_name: downsampled_partial_badclustering_labels
|
222 |
+
data_files:
|
223 |
+
- split: exploratory
|
224 |
+
path: "exploratory/irregular_p30/resampled_1min/bad_partitions/*-labels.parquet"
|
225 |
+
features: *id003
|
226 |
+
- config_name: downsampled_sparse_data
|
227 |
+
data_files:
|
228 |
+
- split: exploratory
|
229 |
+
path: "exploratory/irregular_p90/resampled_1min/*-data.parquet"
|
230 |
+
features: *id001
|
231 |
+
- config_name: downsampled_sparse_labels
|
232 |
+
data_files:
|
233 |
+
- split: exploratory
|
234 |
+
path: "exploratory/irregular_p90/resampled_1min/*-labels.parquet"
|
235 |
+
features: *id002
|
236 |
+
- config_name: downsampled_sparse_badclustering_labels
|
237 |
+
data_files:
|
238 |
+
- split: exploratory
|
239 |
+
path: "exploratory/irregular_p90/resampled_1min/bad_partitions/*-labels.parquet"
|
240 |
+
features: *id003
|
241 |
+
- config_name: raw_complete_data
|
242 |
+
data_files:
|
243 |
+
- split: confirmatory
|
244 |
+
path: "confirmatory/raw/*-data.parquet"
|
245 |
+
features: *id001
|
246 |
+
- config_name: raw_complete_labels
|
247 |
+
data_files:
|
248 |
+
- split: confirmatory
|
249 |
+
path: "confirmatory/raw/*-labels.parquet"
|
250 |
+
features: *id002
|
251 |
+
- config_name: raw_complete_badclustering_labels
|
252 |
+
data_files:
|
253 |
+
- split: confirmatory
|
254 |
+
path: "confirmatory/raw/bad_partitions/*-labels.parquet"
|
255 |
+
features: *id003
|
256 |
+
- config_name: raw_partial_data
|
257 |
+
data_files:
|
258 |
+
- split: confirmatory
|
259 |
+
path: "confirmatory/irregular_p30/raw/*-data.parquet"
|
260 |
+
features: *id001
|
261 |
+
- config_name: raw_partial_labels
|
262 |
+
data_files:
|
263 |
+
- split: confirmatory
|
264 |
+
path: "confirmatory/irregular_p30/raw/*-labels.parquet"
|
265 |
+
features: *id002
|
266 |
+
- config_name: raw_partial_badclustering_labels
|
267 |
+
data_files:
|
268 |
+
- split: confirmatory
|
269 |
+
path: "confirmatory/irregular_p30/raw/bad_partitions/*-labels.parquet"
|
270 |
+
features: *id003
|
271 |
+
- config_name: raw_sparse_data
|
272 |
+
data_files:
|
273 |
+
- split: confirmatory
|
274 |
+
path: "confirmatory/irregular_p90/raw/*-data.parquet"
|
275 |
+
features: *id001
|
276 |
+
- config_name: raw_sparse_labels
|
277 |
+
data_files:
|
278 |
+
- split: confirmatory
|
279 |
+
path: "confirmatory/irregular_p90/raw/*-labels.parquet"
|
280 |
+
features: *id002
|
281 |
+
- config_name: raw_sparse_badclustering_labels
|
282 |
+
data_files:
|
283 |
+
- split: confirmatory
|
284 |
+
path: "confirmatory/irregular_p90/raw/bad_partitions/*-labels.parquet"
|
285 |
+
features: *id003
|
286 |
+
- config_name: correlated_complete_data
|
287 |
+
data_files:
|
288 |
+
- split: confirmatory
|
289 |
+
path: "confirmatory/normal/*-data.parquet"
|
290 |
+
features: *id001
|
291 |
+
- config_name: correlated_complete_labels
|
292 |
+
data_files:
|
293 |
+
- split: confirmatory
|
294 |
+
path: "confirmatory/normal/*-labels.parquet"
|
295 |
+
features: *id002
|
296 |
+
- config_name: correlated_complete_badclustering_labels
|
297 |
+
data_files:
|
298 |
+
- split: confirmatory
|
299 |
+
path: "confirmatory/normal/bad_partitions/*-labels.parquet"
|
300 |
+
features: *id003
|
301 |
+
- config_name: correlated_partial_data
|
302 |
+
data_files:
|
303 |
+
- split: confirmatory
|
304 |
+
path: "confirmatory/irregular_p30/normal/*-data.parquet"
|
305 |
+
features: *id001
|
306 |
+
- config_name: correlated_partial_labels
|
307 |
+
data_files:
|
308 |
+
- split: confirmatory
|
309 |
+
path: "confirmatory/irregular_p30/normal/*-labels.parquet"
|
310 |
+
features: *id002
|
311 |
+
- config_name: correlated_partial_badclustering_labels
|
312 |
+
data_files:
|
313 |
+
- split: confirmatory
|
314 |
+
path: "confirmatory/irregular_p30/normal/bad_partitions/*-labels.parquet"
|
315 |
+
features: *id003
|
316 |
+
- config_name: correlated_sparse_data
|
317 |
+
data_files:
|
318 |
+
- split: confirmatory
|
319 |
+
path: "confirmatory/irregular_p90/normal/*-data.parquet"
|
320 |
+
features: *id001
|
321 |
+
- config_name: correlated_sparse_labels
|
322 |
+
data_files:
|
323 |
+
- split: confirmatory
|
324 |
+
path: "confirmatory/irregular_p90/normal/*-labels.parquet"
|
325 |
+
features: *id002
|
326 |
+
- config_name: correlated_sparse_badclustering_labels
|
327 |
+
data_files:
|
328 |
+
- split: confirmatory
|
329 |
+
path: "confirmatory/irregular_p90/normal/bad_partitions/*-labels.parquet"
|
330 |
+
features: *id003
|
331 |
+
- config_name: nonnormal_complete_data
|
332 |
+
data_files:
|
333 |
+
- split: confirmatory
|
334 |
+
path: "confirmatory/non_normal/*-data.parquet"
|
335 |
+
features: *id001
|
336 |
+
- config_name: nonnormal_complete_labels
|
337 |
+
data_files:
|
338 |
+
- split: confirmatory
|
339 |
+
path: "confirmatory/non_normal/*-labels.parquet"
|
340 |
+
features: *id002
|
341 |
+
- config_name: nonnormal_complete_badclustering_labels
|
342 |
+
data_files:
|
343 |
+
- split: confirmatory
|
344 |
+
path: "confirmatory/non_normal/bad_partitions/*-labels.parquet"
|
345 |
+
features: *id003
|
346 |
+
- config_name: nonnormal_partial_data
|
347 |
+
data_files:
|
348 |
+
- split: confirmatory
|
349 |
+
path: "confirmatory/irregular_p30/non_normal/*-data.parquet"
|
350 |
+
features: *id001
|
351 |
+
- config_name: nonnormal_partial_labels
|
352 |
+
data_files:
|
353 |
+
- split: confirmatory
|
354 |
+
path: "confirmatory/irregular_p30/non_normal/*-labels.parquet"
|
355 |
+
features: *id002
|
356 |
+
- config_name: nonnormal_partial_badclustering_labels
|
357 |
+
data_files:
|
358 |
+
- split: confirmatory
|
359 |
+
path: "confirmatory/irregular_p30/non_normal/bad_partitions/*-labels.parquet"
|
360 |
+
features: *id003
|
361 |
+
- config_name: nonnormal_sparse_data
|
362 |
+
data_files:
|
363 |
+
- split: confirmatory
|
364 |
+
path: "confirmatory/irregular_p90/non_normal/*-data.parquet"
|
365 |
+
features: *id001
|
366 |
+
- config_name: nonnormal_sparse_labels
|
367 |
+
data_files:
|
368 |
+
- split: confirmatory
|
369 |
+
path: "confirmatory/irregular_p90/non_normal/*-labels.parquet"
|
370 |
+
features: *id002
|
371 |
+
- config_name: nonnormal_sparse_badclustering_labels
|
372 |
+
data_files:
|
373 |
+
- split: confirmatory
|
374 |
+
path: "confirmatory/irregular_p90/non_normal/bad_partitions/*-labels.parquet"
|
375 |
+
features: *id003
|
376 |
+
- config_name: downsampled_complete_data
|
377 |
+
data_files:
|
378 |
+
- split: confirmatory
|
379 |
+
path: "confirmatory/resampled_1min/*-data.parquet"
|
380 |
+
features: *id001
|
381 |
+
- config_name: downsampled_complete_labels
|
382 |
+
data_files:
|
383 |
+
- split: confirmatory
|
384 |
+
path: "confirmatory/resampled_1min/*-labels.parquet"
|
385 |
+
features: *id002
|
386 |
+
- config_name: downsampled_complete_badclustering_labels
|
387 |
+
data_files:
|
388 |
+
- split: confirmatory
|
389 |
+
path: "confirmatory/resampled_1min/bad_partitions/*-labels.parquet"
|
390 |
+
features: *id003
|
391 |
+
- config_name: downsampled_partial_data
|
392 |
+
data_files:
|
393 |
+
- split: confirmatory
|
394 |
+
path: "confirmatory/irregular_p30/resampled_1min/*-data.parquet"
|
395 |
+
features: *id001
|
396 |
+
- config_name: downsampled_partial_labels
|
397 |
+
data_files:
|
398 |
+
- split: confirmatory
|
399 |
+
path: "confirmatory/irregular_p30/resampled_1min/*-labels.parquet"
|
400 |
+
features: *id002
|
401 |
+
- config_name: downsampled_partial_badclustering_labels
|
402 |
+
data_files:
|
403 |
+
- split: confirmatory
|
404 |
+
path: "confirmatory/irregular_p30/resampled_1min/bad_partitions/*-labels.parquet"
|
405 |
+
features: *id003
|
406 |
+
- config_name: downsampled_sparse_data
|
407 |
+
data_files:
|
408 |
+
- split: confirmatory
|
409 |
+
path: "confirmatory/irregular_p90/resampled_1min/*-data.parquet"
|
410 |
+
features: *id001
|
411 |
+
- config_name: downsampled_sparse_labels
|
412 |
+
data_files:
|
413 |
+
- split: confirmatory
|
414 |
+
path: "confirmatory/irregular_p90/resampled_1min/*-labels.parquet"
|
415 |
+
features: *id002
|
416 |
+
- config_name: downsampled_sparse_badclustering_labels
|
417 |
+
data_files:
|
418 |
+
- split: confirmatory
|
419 |
+
path: "confirmatory/irregular_p90/resampled_1min/bad_partitions/*-labels.parquet"
|
420 |
+
features: *id003
|
421 |
---
|
422 |
# CSTS - Correlation Structures in Time Series
|
423 |
|
424 |
## Important Notice
|
425 |
+
This dataset is published as a pre-publication release. An accompanying research paper is forthcoming on arXiv.
|
426 |
+
**All usage of this dataset must include proper attribution to the original authors as specified below.**
|
427 |
|
428 |
## Dataset Description
|
429 |
CSTS (**C**orrelation **S**tructures in **T**ime **S**eries) is a comprehensive synthetic benchmarking dataset for
|
430 |
evaluating correlation structure discovery in time series data. The dataset systematically models known correlation
|
431 |
structures between three different time series variates and enables examination of how these structures are affected by
|
432 |
distribution shifting, sparsification, and downsampling. With its controlled properties and ground truth labels,
|
433 |
+
CSTS provides algorithm developers clean benchmark data that bridge the gap between theoretical models
|
434 |
and messy real-world data.
|
435 |
|
436 |
### Key Applications
|
437 |
+
- Evaluating the ability of **time series clustering algorithms** to segment and group segments by correlation structures
|
438 |
+
- Assessing **clustering validation** methods for correlation-based clusters
|
439 |
+
- Investigating how **data preprocessing** affects correlation structure discovery
|
440 |
+
- Establishing **performance thresholds** for high-quality clustering result
|
441 |
|
442 |
### Dataset Structure
|
443 |
CSTS provides **two main splits** (exploratory and confirmatory) with **30 subjects** each, enabling proper statistical validation.
|
444 |
The dataset structure includes:
|
445 |
+
- **12 data variants**: 4 generation stages × 3 completeness levels for each split
|
446 |
+
- **Generation stages**: raw (unstructured data), correlated (normal-distributed data), nonnormal (extreme value and negative binomial distribution shifts), downsampled (1s→1min)
|
447 |
+
- **Completeness levels**: complete (100% of observations), partial (70% of observations), sparse (10% of observations)
|
448 |
|
449 |
### Subjects
|
450 |
+
Each subject contains 100 segments of varying lengths (900-36000) and each segment encodes one of the 23 specific correlation
|
451 |
+
structures. Each subject uses all 23 patterns 4-5 times. For the complete data variants each subject consists of ~1.26 mio
|
452 |
+
observations.
|
453 |
|
454 |
+
Subjects each have the following information, accessible as subsets:
|
455 |
+
- a time series **data file** with three variates (iob, cob, ig) and time stamps (datetime)
|
456 |
- a **label file** specifying the ground truth segmentation and clustering
|
457 |
+
- 67 **bad clustering label files** with controlled degradations (varying numbers of segmentations and/or cluster assignment mistakes) spanning the entire Jaccard Index range [0,1]
|
458 |
|
459 |
### Additional Splits
|
460 |
+
CSTS also includes versions (configured as splits) that allow exploring how cluster and segment counts affect
|
461 |
+
algorithm performance. They follow the same dataset structure (exploratory and confirmatory splits with each 12 data variants subset with each 3 information file subsets).
|
462 |
+
These versions are:
|
463 |
+
- Reduced cluster count: 11 or 6 instead of 23
|
464 |
+
- Reduced segment count: 50 or 25 instead of 100
|
465 |
|
466 |
+
Our accompanying paper provides complete methodological details, baseline findings, and usage guidance. Our
|
467 |
+
GitHub codebase includes the generation, validation and use case code and is configured to automatically load the data.
|
468 |
+
|
469 |
+
## Usage Guidance
|
470 |
+
... coming soon
|
471 |
|
472 |
## Authors
|
473 |
- Isabella Degen, University of Bristol
|
|
|
475 |
- Henry W J Reeve, University of Nanjing
|
476 |
- Kate Robson Brown, University College Dublin
|
477 |
|
|
|
478 |
## Pre-Publication Release Details
|
479 |
- **Release Date:** 29 Apr 2024
|
480 |
- **Version:** 1.0-pre
|
|
|
503 |
Once our paper is published on arXiv, we will update this README with the proper citation information.
|
504 |
**Please check back for updates.**
|
505 |
|
506 |
+
## Acknowledgements
|
507 |
+
... coming soon
|