Anomaly injection for Time Series

Support multiple injections of amplitude shift, distortion , growth change and extreme values of a single data column containing numerical values.

Prerequisites

• The given bash script works for ubuntu 18 or 20
• Clone this repository.

Build

$ sh install.sh

##E xecution

$ python3 inject.py [arguments]

Arguments

-data	-type
stock10k.data 2	amplitude_shift
SAGandStock 2	distortion
	growth_change
	extreme

Remarks

• Data:
  The data has to be in csv style format. To specify a different separator use: -sep "separator"

The file will be saved into the Data/generated folder. The data argument expects the data to be in the Data folder and the second argument is the column starting at 0.

• Parameters:
  The Parameter file specifies the default anomaly parameters. For your own parameters modify the file or use: -typex anomalies your_parameter_file.

• Arguments:
  All the arguments try to match if only the beginning is given e.g., a insteaf of amplitude_shift. One can use multiple anomalies at once e.g., -type a,a,d,d

• Plot and Save commands
  -withoutlegend
  -save filename
  -anomalydetails
  -plotoff

Examples:

$ python3 inject.py -data stock10k.data 2 -type  amplitude_shift,distortion -anomalydetails

$ python3 inject.py -data stock10k.data 2 -type a,d 

$ python3 inject.py -data stock10k.data 2 -type a -save output 

$ python3 inject.py -data SAGandStock  2  -save "2shifts" -typex a,a Parameters -plotoff