get_group_size = lambda df: df.reset_index().groupby("user_id")['user_id'].count()

train_groups = get_group_size(xtrain)
test_groups = get_group_size(xtest)

print(sum(train_groups) , sum(test_groups))
#(4764372, 100000)

model = LGBMRanker(objective="lambdarank")
model.fit(xtrain,ytrain,group=train_groups,eval_set=[(xtest,ytest)],eval_group=[test_groups],eval_metric=['ndcg'])

#....
# [97] valid_0's ndcg@1: 0.900624 valid_0's ndcg@2: 0.900015 valid_0's ndcg@3: 0.892648 valid_0's ndcg@4: 0.891373 valid_0's ndcg@5: 0.886585
# [98] valid_0's ndcg@1: 0.900624 valid_0's ndcg@2: 0.900015 valid_0's ndcg@3: 0.892648 valid_0's ndcg@4: 0.891895 valid_0's ndcg@5: 0.886632
# [99] valid_0's ndcg@1: 0.900624 valid_0's ndcg@2: 0.901216 valid_0's ndcg@3: 0.892839 valid_0's ndcg@4: 0.892053 valid_0's ndcg@5: 0.88677
# [100] valid_0's ndcg@1: 0.900624 valid_0's ndcg@2: 0.901216 valid_0's ndcg@3: 0.892839 valid_0's ndcg@4: 0.892053 valid_0's ndcg@5: 0.886363