Should we consider a new design to no algorithm engineer modification needed?

[joshua-xia]

What would you like to be added:
a new design let algorithm engineer need NOT modify the code according to the parameters input

Why is this needed:
reduce algorithm engineer 's work

Without this feature, how does current nni work：
current nni need modify the code according to the parameters

Components that may involve changes:
the pipeline design

Brief description of your proposal if any:
Not yet decide

[xuehui1991]

Hello @joshua-xia, could you give more detail about your proposal? Could you give some examples? Thank you.

[joshua-xia]

I am trying to write a prototype. you know to use the current NNI, the algorithm engineer have to modify the model code according to the parameters input, and report the results using the SDK API, I wonder if there is a design/implement reduce this kind of modification

[joshua-xia]

for example:

I have to write the following code to get model:

def get_model(PARAMS):
'''Get model according to parameters'''
if PARAMS.get('_name') == 'GradientBoostingClassifier':
model = ensemble.GradientBoostingClassifier()
model.loss = PARAMS.get('loss')
model.learning_rate = PARAMS.get('learning_rate')
model.n_estimators = PARAMS.get('n_estimators')
model.criterion = PARAMS.get('criterion')
elif PARAMS.get('_name') == 'SVM':
import sklearn.svm.SVC
model = sklearn.svm.SVC()
model.C = PARAMS.get('C')
model.keral = PARAMS.get('keral')
model.degree = PARAMS.get('degree')
model.gamma = PARAMS.get('gamma')
model.coef0 = PARAMS.get('coef0')
elif PARAMS.get('_name') == 'RF':
....

Can we provide API in NNI SDK such as: nni.get_next_model() to get model directly?

[scarlett2018]

Thanks @joshua-xia for raising the feature request. This is not in our recent roadmap, we will leave this open in case any other users have the interests or raise any proposals.