[GSK-989] doc related issues (xhr)

[rabah-khalek]

rabah-khalek:

• [ ] It’s not clear that classification_label are computed automatically, and which element of an array is the lowest/highest probability
• [ ] For optional argument in dataset and model, we should describe what the optional parameter are (example : classification_label)

@kevinmessiaen :

• [X] DataQuality can’t be resolved from giskard
  • [X] if he skipped step 4, he tried to look for DataQuality in our doc, we need to put some link between step 4 and 5
• [X] We should make sure that our snippet work without relying on previous snippet
• [ ] He’s not sure wether to use the decorator or the class to create a custom test
  • [X] \~he preferred to use a decorator\~
• [X] We should not put in the documentation the custom test part at first. It should more hidden. It’s hard for the user to understand what custom means
• [ ] He’s not sure about what are the parameters of run and think that we can only test the same model
• [X] doc: test_f1(actual_dataset,…) —> test_f1(dataset,…)
• [X] This copied snippet gave him an error: ⇒ it’s not the latest version of GitBook doc that has been copied 😕

print(f"Result: {passed}")print(f"F1: {results['f1'].passed} {results['f1'].metric}")print(f"DataQuality: {results['quality'].passed} {results['quality'].metric}")

• [ ] we should work on documenting the different input scenario to test-suite, it was difficult for him to know what to do without hints from JM: we should have a section title in guide called : “Compare 2 models with a suite”, “compare two 2 dataset with a suite”
• [ ] In the section Apply your slicing function to a dataset#,  we need to say we can convert df to pandas df
• [X] In the chaining part, what is the "WHY?" of the section "chain slices and transformation". The guide should be functional
• [X] In the chaining part, why do we hava
  • [X] the method add_transformation instead of transform
  • [X] the method add_slicing_function instead of slice
  • [X] the process method. What's the use?

_{From SyncLinear.com | GSK-989}

[rabah-khalek]

@jmsquare : is there a way to enlarge "warnings" and "hints" on sphinx?

[rabah-khalek]

put:

In the case of binary classification, an array of probabilities   is also accepted. Make sure that the probability provided is for the first label provided in classification_labels.

next to classification_labels.

Stress on the fact that classification_labels and classification_threshold are relevant only for classification

use iris as example for snippets in model and dataset (https://rasbt.github.io/mlxtend/user_guide/data/iris_data/):

iris = datasets.load_iris()
df_iris = pd.DataFrame(iris.data,columns=iris.feature_names)

d = {0: 'setosa', 1: 'versicolor', 2: 'virginica'}

Classfication_labels: Type: dict or a list The mapping of your classification labels. It’s a dictionnary where the

• Keys are values in the target variable of the dataset you upload in Giskard
• Values are human-readable labels

If the keys are already human readable you can simply use put your classification labels as list of string

[rabah-khalek]

add tabs for each of:

• multilabel classification
• binary classfication
• regression

in the guides and only highlight relevant parameters

[andreybavt]

@rabah-khalek-khalek is this card still in progress?

[rabah-khalek]

What needs to be implemented was already implemented here.