I've got a question about finding the predictive value of certain attributes.
In my question it is suggested that I transform my attributes to binary classes and then apply "decision stump" to find out the predictive value of each attribute. How do I do this?
I checked out this question But that's not really what I mean.
Thanks in advance, Rope.
You can find the most predictive attributes using the methods found under the Select Attributes tab in Weka's Explorer.
Yeah, the Select Attributes tab in Weka analyzes your attributes and ranks which ones provide the most information gain. Under Attribute Evaluator, choose InfoGainAttributeEval and choose Ranker for search method. It works quite well in my experience.
Related
I have a flask API that handles ML models and I want to get metrics based on each model but the way we grab each model is based on the path. ex: https://api.someapi.com/cmd/<model-name> is it possible to grab metrics per model name? I was reading the docs and I guess we could hack something together with labels but that doesn't seem like the right solution. Has anyone run into this problem? I am using prometheus-flask-exporter.
labels is always the right solution in Prometheus' to differentiate between otherwise identical metrics.
The "trick" is in identifying the unique metrics and the way to determine these is to ask "Will we ever want to (dynamically) query metrics by all or by some label values?". Corollary: what measurements of your models will you want to compare across models?
Without knowing more about your solution, you may have metrics corresponding to model invocations, failures etc. and you would differentiate between e.g. model invocations by applying at least one label.
In your case, perhaps a label key would be model_name and you'd use the path parameter <model-name> as the value.
It's unclear whether your paths are statically or dynamically generated but it makes no difference to the metrics and labels; just how you apply them.
I have a page where I am listing some entities and providing an interface with multiple filtering options. For simplifying the question, Let's say I am listing various movies on that page. So de-normalized row should look like this for a single movie entry;
producer_id: Partition Key - (e.g: PRODUCER#213141)
movie_id: Sort Key - (e.g: MOVIE#887347)
producer_name: (e.g: "Warner Bros")
movie_name: (e.g: "Harry Potter")
status: (e.g: "ON_SHOW")
publish_date: (e.g: "2020.01.01")
type: (e.g: "fantasy")
language:(e.g: "English")
I want to enable filtering by using a composite attribute used as a GSI secondary key. My composite attribute would look like something like this;
GS1SK: "harry_potter#2020.01.01#fantasy#English#ON_SHOW"
And the partition key for this secondary index would be simply the producer_id.
So let's say any user comes to the page and wants to filter out the movies with the given filtering options. Such an access pattern example would be;
Get all of the sci-fi movies produced by Warner Bros and has
the status ON_SHOW.
However, the problem starts here. Since the combined attributes are not hierarchical, the composite attribute can't be used for this scenario. Because in the above access pattern example, the user had the option to not specify any date, simply wanted to get results for any date range. But when you look at the structure of the composite attribute, it is impossible to not specify the date range for filter other attributes in advance such as movie type or movie language.
I know the DynamoDB is not the best fit for such complex querying, however, I think providing filtering options in a listing page is a really typical scenario that even the simplest products should provide. My question is, what kind of approach should I use to satisfy this filtering needs.
Maybe I am getting the idea wrong behind composite attributes?
Should I use filtering expressions and there is no way to do such advanced filtering with composite GSIs?
Maybe for such scenarios, I should be considering Elastic Search or AWS Athena services?
I will need to provide even more filtering options in the other pages of my application, such as filtering listed users with their demographic information. Do you think I should be considering migrating RDBS rather than using a NO-SQL database?
What I really want to do is to provide these filtering features without any filtering expressions to reduce RCU usage, and increase the efficiency in my queries. I would appreciate any kind of help and advice. Thanks.
It sounds like you understand composite attributes perfectly and have a solid grasp of your options. You've stumbled across one of the weaknesses of DynamoDB. It's challenging to support this kind of ad-hoc search functionality with DynamoDB.
I've seen this problem solved using a tool like Elastic Search (your option #3). A common pattern is to enable DynamoDB streams, which can be used to update the Elastic Search index. A bit more infrastructure to set up, but the search capabilities would be much more flexible than what you get with DynamoDB alone.
I am trying to change the dropdown options for Reason and Severity in Cases. Does anyone know how to do that?
You change the values within the selector in a variety of ways, I would be mindful to update defaulting logic as well as paying close attention to how Acumatica responds to the current values in processing.
Extending PXStringListAttribute with additional values.
Replace Items within PXStringListAttribute with custom values.
https://riptutorial.com/acumatica/example/29102/modifying-marital-statuses
I'd like to display the contents of a database table in a more interesting manner than just a table; different positioning, fonts, sizes, colours etc.
Is there a region type or plugin that can achieve that?
I tried the cards view but I it's still very limited (for example a minimum of 2 cards per row when I only need 1) Or maybe I should deploy it as a classic report and then use javascript to rearrange everything? edit: I also considered rendering the page with pl/sql and htp, but that seems overly complicated...
As I'm new to APEX, any input would be much appreciated!
You could do it as a classic report and change the template under the attributes to something else.
What I ended up doing was using an interactive report with a 'detail view'. That way you can use html+css and reference the data from the columns by using #column_name#. I'm really satisfied with it.
However the above answers are also very interesting! I definitely learned a lot experimenting with them.
I have custom model with different data types like string (file path) or double which should be edited using combobox with a few items.
It is not clear for me why delegates should be applied to views and not to models...
So, should I pass some kind of flag from my model and then use different delegates for those items according to those flags?
What is the best practice to make it?
EDIT: I'll try to clarify my question:
I have model with the map (key - value (structure that can contain different types like QVariant)) and it's necessary to set different delegates for each row of my, say, QTableView.
What is the best way to pass some "flag" for every item from my model and then handle this flag to set appropriate delegate for the given row?
EDIT2:
This model-view pair is for storage and editing software options with different types.
From the docs:
Unlike the Model-View-Controller pattern, the model/view design does not include a completely separate component for managing interaction with the user.
Delegates are supposed to tackle the "how" in "how should users interact with my data" (that's why I highlighted "interaction").
For your case, that very "double" field you provided, depending on it's interpretation, we could use a line edit (eg exact toleration), spinbox, or even some sort of color select (interpret the value as a color). Even more, one could use a line edit with some sort of color scale for the widget to make it more clear what consequence that value may have.
Correct way? They're tools, not one better than other but rather "one to tackle a specific problem". Can't tell what's the correct way from the info provided.
I suggest re-asking the question with much more info if you still have doubts.