177 finish kalman tutorial #178
Conversation
|
|
||
| You can see that there is now a KnowledgeRep plugin, Kalman Filter: this plugin will see the low-level data frame and use it to build up higher level information. | ||
| In this case, a list of telemetry points that have violated their Kalman filter prediction for a time step. | ||
| There is also a Learners plugin, csv output: normally a learner would use both the low-level data and the higher level information from the Knowledge Representation to TODO. |
There was a problem hiding this comment.
Is the TODO still here by design?
| The Kalman Plugin [kalman_plugin.py](../onair/plugins/kalman/kalman_plugin.py) is an [`AIPlugin`](../onair/src/ai_components/ai_plugin_abstract/ai_plugin.py) which requires two functions: `update` and `render_reasoning`. | ||
| `AIPlugin` also defines an `__init__` function which stores the plugin name and a list of names of the telemetry points in the low level data (called "headers"). |
There was a problem hiding this comment.
We should mention inheritance here. It does say 'is an' but specifically mentioning inheritance would make it clear. We should also state that AIPlugin is an abstract class as this is the first mention of AIPlugin in the tutorial. This helps explain how a plugin is required to implement the functions and gain the base __init__.
| ComplexPluginDict = {} | ||
| ``` | ||
|
|
||
| Note: you can set plugins to be blank or the 'generic' empty plugin. |
There was a problem hiding this comment.
I do not understand the note.
|
|
||
| Note: you can set plugins to be blank or the 'generic' empty plugin. | ||
|
|
||
| You can see that there is now a KnowledgeRep plugin, Kalman Filter: this plugin will see the low-level data frame and use it to build up higher level information. |
There was a problem hiding this comment.
I suggest changing 'now' to 'currently', but only because as this is the tutorial using of the word 'now' makes it feel as though we should have just changed something (although there was no direction to have done so).
| When it drops to 2.0, "CURRENT" is added to the output from the Kalman plugin. | ||
| This happens again when the current jumps again from 3.0 to 5.0. | ||
|
|
||
| Note: not every telemetry point flagged by the Kalman filter is idicative of a fault, some are false positives. |
There was a problem hiding this comment.
Should we provide a rationale as to why this note is true?
|
|
||
| ## Write a New Plugin | ||
|
|
||
| Now we are going to write a new plugin to "scream" every time a tlemetry point has been marked as broken by the Kalman plugin. |
| Next, the plugin checks the Kalman plugins rendered output (`high_level_data['vehicle_rep']['Kalman Filter']` is the list returned by the Kalman plugin's `render_reasoning` function). | ||
| If something is there, the scream plugin will, well scream about it. | ||
|
|
||
| When you run your scream example, you should now see the following amoung the output: |
| First we check that the high level data does in fact have an entry named "Kalman Filter", which is the name of the Kalman plugin in the config file. | ||
|
|
||
| Next, the plugin checks the Kalman plugins rendered output (`high_level_data['vehicle_rep']['Kalman Filter']` is the list returned by the Kalman plugin's `render_reasoning` function). | ||
| If something is there, the scream plugin will, well scream about it. |
There was a problem hiding this comment.
I appreciate the funny! You may want a comma after well, or maybe ellipsis it, well... or, if not keeping the joke, just remove the well entirely.
| pass | ||
| ``` | ||
|
|
||
| We do everything in the `update` function since scream doesn't actually have anything useful to pass on to other plugins. |
There was a problem hiding this comment.
Is this because a Complex Reasoner plugin does its update and render_reasoning in succession? We may want to offer more of an explanation of why this was chosen for the tutorial to allay any misunderstanding.
Updates the Kalman filter tutorial example. Closes #177
The tutorial now shows what the Kalman filter plugin is doing, the data flow of the Kalman example, and steps the user through writing their own plugin to scream when something goes wrong (in the data).