fix
Browse files
README.md
CHANGED
@@ -8,16 +8,19 @@ size_categories:
|
|
8 |
---
|
9 |
|
10 |
# Pushing the Limits of Pre-training for Time Series Forecasting in the CloudOps Domain
|
|
|
11 |
[Paper](https://arxiv.org/abs/2310.05063) | [Code](https://github.com/SalesforceAIResearch/pretrain-time-series-cloudops)
|
12 |
|
13 |
Datasets accompanying the paper "Pushing the Limits of Pre-training for Time Series Forecasting in the CloudOps Domain".
|
14 |
|
|
|
15 |
```python
|
16 |
from datasets import load_dataset
|
17 |
|
18 |
dataset = load_dataset('Salesforce/cloudops_tsf', 'azure_vm_traces_2017')
|
19 |
```
|
20 |
|
|
|
21 |
### azure_vm_traces_2017
|
22 |
```python
|
23 |
DatasetDict({
|
@@ -60,31 +63,57 @@ DatasetDict({
|
|
60 |
})
|
61 |
```
|
62 |
|
63 |
-
|
64 |
```python
|
65 |
from datasets import load_dataset_builder
|
66 |
|
67 |
config = load_dataset_builder('Salesforce/cloudops_tsf', 'azure_vm_traces_2017').config
|
68 |
-
config
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
```
|
|
|
|
|
|
|
70 |
|
71 |
## Acknowledgements
|
72 |
The datasets were processed from the following original sources. Please cite the original sources if you use the datasets.
|
73 |
* Azure VM Traces 2017
|
74 |
-
* Bianchini. Resource central: Understanding and predicting workloads for improved resource
|
75 |
-
management in large cloud platforms. In Proceedings of the 26th Symposium on Operating Systems
|
76 |
-
Principles, pp. 153–167, 2017.
|
77 |
* https://github.com/Azure/AzurePublicDataset
|
78 |
|
79 |
* Borg Cluster Data 2011
|
80 |
-
* John Wilkes. More Google cluster data. Google research blog, November 2011. Posted at http
|
81 |
-
//googleresearch.blogspot.com/2011/11/more-google-cluster-data.html.
|
82 |
* https://github.com/google/cluster-data
|
83 |
|
84 |
* Alibaba Cluster Trace 2018
|
85 |
-
* Jing Guo, Zihao Chang, Sa Wang, Haiyang Ding, Yihui Feng, Liang Mao, and Yungang Bao. Who
|
86 |
-
limits the resource efficiency of my datacenter: An analysis of alibaba datacenter traces. In
|
87 |
-
Proceedings of the International Symposium on Quality of Service, pp. 1–10, 2019.
|
88 |
* https://github.com/alibaba/clusterdata
|
89 |
|
90 |
## Citation
|
|
|
8 |
---
|
9 |
|
10 |
# Pushing the Limits of Pre-training for Time Series Forecasting in the CloudOps Domain
|
11 |
+
|
12 |
[Paper](https://arxiv.org/abs/2310.05063) | [Code](https://github.com/SalesforceAIResearch/pretrain-time-series-cloudops)
|
13 |
|
14 |
Datasets accompanying the paper "Pushing the Limits of Pre-training for Time Series Forecasting in the CloudOps Domain".
|
15 |
|
16 |
+
## Quick Start
|
17 |
```python
|
18 |
from datasets import load_dataset
|
19 |
|
20 |
dataset = load_dataset('Salesforce/cloudops_tsf', 'azure_vm_traces_2017')
|
21 |
```
|
22 |
|
23 |
+
## Available Datasets
|
24 |
### azure_vm_traces_2017
|
25 |
```python
|
26 |
DatasetDict({
|
|
|
63 |
})
|
64 |
```
|
65 |
|
66 |
+
## Dataset Config
|
67 |
```python
|
68 |
from datasets import load_dataset_builder
|
69 |
|
70 |
config = load_dataset_builder('Salesforce/cloudops_tsf', 'azure_vm_traces_2017').config
|
71 |
+
print(config)
|
72 |
+
|
73 |
+
CloudOpsTSFConfig(
|
74 |
+
name='alibaba_cluster_trace_2018',
|
75 |
+
version=1.0.0,
|
76 |
+
data_dir=None,
|
77 |
+
data_files=None,
|
78 |
+
description='',
|
79 |
+
prediction_length=48,
|
80 |
+
freq='5T',
|
81 |
+
stride=48,
|
82 |
+
univariate=False,
|
83 |
+
multivariate=True,
|
84 |
+
optional_fields=('feat_static_cat', 'past_feat_dynamic_real'),
|
85 |
+
rolling_evaluations=12,
|
86 |
+
test_split_date=Period('2018-01-08 11:55', '5T'),
|
87 |
+
_feat_static_cat_cardinalities={
|
88 |
+
'pretrain': (
|
89 |
+
('container_id', 64457),
|
90 |
+
('app_du',9484)),
|
91 |
+
'train_test': (
|
92 |
+
('container_id', 6048),
|
93 |
+
('app_du', 1292)
|
94 |
+
)
|
95 |
+
},
|
96 |
+
target_dim=2,
|
97 |
+
feat_static_real_dim=0,
|
98 |
+
past_feat_dynamic_real_dim=6
|
99 |
+
)
|
100 |
```
|
101 |
+
```test_split_date``` is provided to achieve the same train-test split as given in the paper.
|
102 |
+
This is essentially the date/time of ```rolling_evaluations * prediction_length``` time steps before the last time step in the dataset.
|
103 |
+
Note that the pre-training dataset includes the test region, and thus should also be filtered before usage.
|
104 |
|
105 |
## Acknowledgements
|
106 |
The datasets were processed from the following original sources. Please cite the original sources if you use the datasets.
|
107 |
* Azure VM Traces 2017
|
108 |
+
* Bianchini. Resource central: Understanding and predicting workloads for improved resource management in large cloud platforms. In Proceedings of the 26th Symposium on Operating Systems Principles, pp. 153–167, 2017.
|
|
|
|
|
109 |
* https://github.com/Azure/AzurePublicDataset
|
110 |
|
111 |
* Borg Cluster Data 2011
|
112 |
+
* John Wilkes. More Google cluster data. Google research blog, November 2011. Posted at http://googleresearch.blogspot.com/2011/11/more-google-cluster-data.html.
|
|
|
113 |
* https://github.com/google/cluster-data
|
114 |
|
115 |
* Alibaba Cluster Trace 2018
|
116 |
+
* Jing Guo, Zihao Chang, Sa Wang, Haiyang Ding, Yihui Feng, Liang Mao, and Yungang Bao. Who limits the resource efficiency of my datacenter: An analysis of alibaba datacenter traces. In Proceedings of the International Symposium on Quality of Service, pp. 1–10, 2019.
|
|
|
|
|
117 |
* https://github.com/alibaba/clusterdata
|
118 |
|
119 |
## Citation
|