1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
|
# Model Visualization
## Textual Output
In low- to medium-complexity applications, the simplest way of examining the
performance models generated by dfatool is textual output: `--show-model=param`.
This is implemented for most modeling methods.
## Graphical Output via dot(1)
`--export-dot PREFIX` exports the generated performance models for each pair of
name (state/transition/…) and performance attribute to
`PREFIX(name)-(attribute).dot`. The dot(1) program is capable of transforming
those into images. For instance, if feh(1) is available, a visual model is
accessible via `dot -Tpng filename.dot | feh -`.
In case of regression forests (XGBoost), dfatool exports the individual trees to
`PREFIX(name)-(attribute).(index).dot`.
## Plotting Model Predictions for Individual Configurations
`--plot-param name:attribute:parameter` displays both raw readings (as points,
see [[analysis-visual.md]]) and the corresponding performance model (as lines).
The plot is saved to (name)-(attribute)-(parameter).pdf and shown interactively
unless `--non-interactive` has been specified.
## JSON Export
Use `--export-json FILENAME` to export the performance model to FILENAME.
The model for NAME ATTRIBUTE is located in .name.NAME.ATTRIBUTE.modelFunction.
If it is a list, it is an XGB forest. Otherwise, it is an CART/LMT/RMT element
described by the `type` key.
### CART/LMT scalar split node
```
{
"value": 123,
"type": "scalarSplit",
"paramIndex": 0,
"threshold": 24,
"left": …,
"right": …
}
```
The model for `param[paramIndex] <= threshold` is located in `left`, the model
for `param[paramIndex] > threshold` is located in `right`. `value` is a
static model that serves as fall-back if `param[paramIndex]` is undefined.
### RMT categorical split node
```
{
"value": 123,
"type": "split",
"paramIndex": 0,
"child": {
"y": …,
"n": …,
}
}
```
Given some (usually not numeric) `value`, the model for `param[paramIndex] ==
value` is located in `child[value]`. In this example, the node describes a
boolean Kconfig feature, so there are just two children: `y` (feature enabled)
and `n` (feature disabled). Again, `value` refers to a static fall-back model.
### ULS (least-squares regresion) node
```
{
"value": 123,
"type": "analytic",
"functionStr": "0 + regression_arg(0) + regression_arg(1) * np.log(parameter(bitrate))",
"regressionModel": [674574496.5381737, -51732985.15977712]
}
```
The model function is `674574496.5381737 - 51732985.15977712 * log(bitrate)`.
`value` is a static model that serves as fall-back if `bitrate` is undefined.
### Static node
```
{
"value": 123,
"type": "static"
}
```
|
