Parameter Package

Table 1 describes the parameters.

**Table 1** Parameters
Parameter	Description	Input/Output	Type	Value Range	Default Value
detect_points	Number of source data points to be detected	Input	int	[1, +∞)	120
valid_factor_default	Default parameter for controlling model sensitivity	Input	float	(0.0, +∞)	1.0
detection_window	Size of the sliding window for detection	Input	int	[1, +∞)	5
slide_window	Step of the sliding window	Input	int	[2, total number of drives to be detected]	5
train_model_min	Minimum radius of the DBSCAN algorithm	Input	float	(0.0, +∞)	0.3
log_file	Log file path.	Input	String	-	/var/log/smartmaintainkit.log

The slow drive detection effect is mainly determined by two indicators.

Width and step of the sliding window: All source data is divided into sliding windows, and the value of each sliding window is packed into a test point.
1. detection_window specifies the width of the sliding window.
2. slide_window specifies the sliding step of the window.
  Example:
  The default parameters are used for invoking the detection interface as follows. However, the latency of drives in the dataset is low, and slow drives cannot be detected.
```
slow_detect(r'slow_test.data', r'result.data')
```
  Solution: Adjust the width of the sliding window through detection_window and narrow the step of the sliding window through slide_window to collect latency data in a more delicate fashion.
```
param = {'detection_window': 4, 'slide_window': 2} 
slow_detect(r'slow_test.data', r'result.data', param)
```
Slow drive detection precision: This metric is determined by the radius of the DBSCAN algorithm, and the radius is determined by valid_factor_default and train_model_min.
1. valid_factor_default is a factor that is calculated with the value of each detection point to obtain the radius of the DBSCAN algorithm. A larger value of valid_factor_default indicates a larger radius of the DBSCAN algorithm and a lower precision of slow drive detection.
2. The radius of the DBSCAN algorithm must be greater than 0. Therefore, a minimum value is set to control the radius of the DBSCAN algorithm calculated based on valid_factor_default. If "WARNING: Model is too low, set model with train_model_min" is displayed in the log, the radius calculated using valid_factor_default is less than the value of train_model_min. In this case, the value of train_model_min is used as the radius of the DBSCAN algorithm.
  Example:
  Slow drives cannot be detected when default parameters are used due to narrow latency differences between slow drives and normal latency.
```
slow_detect(r'slow_test.data', r'result.data')
```
  Solution: Reduce the value of valid_factor_default.
```
param = {'valid_factor_default': 0.5} 
slow_detect(r'slow_test.data', r'result.data', param)
```
  If "WARNING: Model is too low, set model with train_model_min" is displayed in the log, the minimum radius specified by train_model_min is reached. In this case, decrease the value of train_model_min to improve detection precision.
```
param = {'valid_factor_default': 0.5, 'slide_window': 0.1} 
slow_detect(r'slow_test.data', r'result.data', param)
```

Parent topic: Interfaces and Parameters