--- license: other license_name: yahoo-dataset license_link: LICENSE.md pretty_name: Yahoo! Synthetic and real time-series with labeled anomalies, version 1.0 extra_gated_prompt: Yahoo Dataset [Terms of Use](https://huggingface.co/datasets/YahooResearch/ydata-labeled-time-series-anomalies-v1_0/blob/main/LICENSE.md) extra_gated_fields: Name: text Email: text Organization: text Country: country I agree to use this dataset for non-commercial use: checkbox I accept the Terms of Use linked above: checkbox --- # Yahoo! Synthetic and real time-series with labeled anomalies, version 1.0 This dataset contains 371 files: ``` A1Benchmark/real_(int).csv A2Benchmark/synthetic_(int).csv A3Benchmark/A3Benchmark-TS(int).csv A4Benchmark/A4Benchmark-TS(int).csv ``` A1Benchmark is based on the real production traffic to some of the Yahoo! properties. The other 3 benchmarks are based on synthetic time-series. A2 and A3 Benchmarks include outliers, while the A4Benchmark includes change-point anomalies. The bechmarks based on real-data have property and geos removed. Fields in each data file are delimited with `,` characters. The content of the files are as follows: ## Real and synthetic time series `A[1-2]Benchmark/[real,synthetic]_(int).csv` contains real and synthetic time-series with labeled anomalies. The synthetic data set contains time-series with random seasonality, trend and noise. The outliers in the synthetic dataset are inserted at random positions. Note that the timestamps of the A1Benchmark are replaced by integers with the increment of 1, where each data-point represents 1 hour worth of data. The anomalies in A1Benchmark are marked by humans and therefore may not be consistent, therefore during benchmarking A1Benchmark is best used to measure recall. The fields are: 0. `timestamp` 1. `value` 2. `is_anomaly` The `is_anomaly` field is a boolean indicating if the current value at a given timestamp is considered an anomaly. Snippet: ``` 1,83,0 2,605,0 3,181,0 4,37,0 5,45,1 ``` ## Synthetic time series `A[3-4]Benchmark/A[3-4]Benchmark-TS(int).csv` contain synthetic time-series. The A3Benchmark only contains outliers while the A4Benchmark also contains the anomalies that are marked as change-points. The synthetic time-series have varying noise and trends with three pre-specified seasonalities. The anomalies in the synthetic time-series are inserted at random positions. The fields are: 0. `timestamps`: the UNIX timestamp marks every hour (the data is hourly sampled) 1. `value`: the value of time series at this timestamp 2. `anomaly`: 1 if this stamp is an outlier 3. `changepoint`: 1 if this stamp is a change point 4. `trend`: the additive trend value for this timestamp 5. `noise`: the additive noise value for this timestamp 6. `seasonality1`: the 12-hour seasonality value 7. `seasonality2`: the daily seasonality value 8. `seasonality3`: the weekly seasonality value Snippet: ``` 1422237600,4333.43325915382,0,0,4599,1.81512268926974,-190.958601534458,-128.864580083128,52.4413180821374 1422241200,4316.14322293657,0,0,4602,-14.6572208523525,-220.5,-105.217489040563,54.5179328294825 1422244800,4403.20006523115,0,0,4605,7.04036744752875,-190.958601534463,-74.3999999999969,56.5182993180795 1422248400,4531.20632084718,0,0,4608,13.5289749039305,-110.250000000009,-38.5122739112586,58.4396198545142 1422252000,4967.50678185938,1,0,4911,-3.7724254388993,-6.91288625182698e-12,-2.33251127952801e-12,60.279207 ```