How to let deployed classifier return probabilities?

[caojianneng]

Description of the problem:

#Let the random-forest model be: rf_model

from kale.common.serveutils import serve
kfserver = serve(rf_model) #model is deployed

#prepare data for prediction
data = [row.tolist() for _, row in train_df[predictor_var].head(10).iterrows()]
data_json = json.dumps({"instances": data})

#prediciton:
pred = kfserver.predict(data_json)

Question 1: The returned pred is class labels: 0/1. How to return probabilities?

I tried the following way after studying: kale.common.serveutils.predict
#let HOST be the host name of deployed model
#let the URL of calling the deployed model be: http://xxx:predict
headers = {"content-type": "application/json", "Host": HOST}
pred_2 = requests.post(url = URL,
data=data_json,
headers=headers)

Question 2: But not clear, where to set parameter, so pred_2 will return probabilities?