Cold Star Problem New User

Nobody has claimed this yet.

I want to recommend items to user 4 with user feature ({"user_id":"4","user_city":"Z"}) who is not in training set. It's a cold star problem.
I have read similar question and answer but it is not clear as you said that set all ID 0s. which IDs ? and Where?
Please suggest what changes should i need to do in the given code and data

`import numpy as np
import json
from itertools import islice

news_features = [{"news_id":"1","car_model":"A"},
{"news_id":"2","car_model":"A"},
{"news_id":"3","car_model":"B"},
{"news_id":"4","car_model":"C"}]

user_features = [{"user_id":"1","user_city":"X"},
{"user_id":"2","user_city":"Y"},
{"user_id":"3","user_city":"Z"}]

ratings = [{"user_id":"1", "news_id":"1", "rating":"1"},
{"user_id":"1", "news_id":"2", "rating":"5"},
{"user_id":"1", "news_id":"3", "rating":"5"},
{"user_id":"1", "news_id":"4", "rating":"1"},
{"user_id":"2", "news_id":"3", "rating":"4"},
{"user_id":"2", "news_id":"1", "rating":"5"},
{"user_id":"3", "news_id":"1", "rating":"5"},
{"user_id":"3", "news_id":"3", "rating":"1"}]

from lightfm.data import Dataset
dataset = Dataset()

dataset.fit(users=(x['user_id'] for x in user_features),
items=(x['news_id'] for x in news_features),
item_features=(x['car_model'] for x in news_features),
user_features=(x['user_city'] for x in user_features))

num_users, num_items = dataset.interactions_shape()

(interactions, weights) = dataset.build_interactions(((x['user_id'], x['news_id'], int(x['rating']))
for x in ratings))

item_features = dataset.build_item_features(((x['news_id'], [x['car_model']])
for x in news_features))

user_features = dataset.build_user_features(((x['user_id'], [x['user_city']])
for x in user_features))

from lightfm import LightFM
model = LightFM(loss='warp')
model.fit(interactions, item_features=item_features, user_features=user_features, epochs=10, num_threads=4)

### Recommended items to user 3
user_features_test = dataset.build_user_features(((x['user_id'], [x['user_city']])
for x in [{"user_id":"3","user_city":"Z"}]))
scores = model.predict(2, np.arange(num_items), user_features=user_features_test)`

Start by reading the Dataset.fit, build_user_features, and model.predict calls shown in the issue, then compare how the training users and target user 4 are represented. Check the LightFM documentation and relevant examples for recommendations involving users absent from the training interactions. Done means the required data and code changes for this cold-start case are clearly identified and verified.

Written by the indexing model from the issue text.