Score_mode avg returns 1 for all documents - elasticsearch

Question

I'm using function score with score_mode avg and boost_mode replace.
And according to documentation I'm expecting the query function score to be overriden by the function score filters (because I'm using boost_mode replace).
This works as expected for the sum and multiply, but not for avg (I'm aware that the average in function score is a weighted average)
When I apply this function_score all the documents get a score of 1.
How can this happen?
GET kibana_sample_data_ecommerce/_search
{
"_source": {
"includes": ["customer_last_name", "customer_first_name", "customer_gender"]
},
"size": 10,
"query": {
"function_score": {
"functions": [
{
"filter": { "match": { "customer_last_name": "Cook" } },
"weight": 2
},
{
"filter": { "match": { "customer_first_name": "Jackson" } },
"weight": 4
},
{
"filter": { "match": { "customer_gender" : "MALE"} },
"weight": 8
}
],
"score_mode": "avg",
"boost_mode": "replace"
}
}
}

Answer 1

So this is a bit weird, but the link provided by #jzzfs is already pretty close. The average mode of the function score query provides a weighted average, which causes this effect:
In case score_mode is set to avg the individual scores will be combined by a weighted average. For example, if two functions return score 1 and 2 and their respective weights are 3 and 4, then their scores will be combined as (1*3+2*4)/(3+4) and not (1*3+2*4)/2.
In addition, it's important to note that due to this, function scores whose filters don't match the current document have no effect on the average score, rather then reducing it. In your example, this means that if a document only matches by having a MALE customer, it will have a score of 8, but since it's weighted it'll actually have a score of (1*8)/8 = 1. If it's a MALE with the first name Jackson, the score again will be (1*8 + 1*4)/(8+4)=1. This can easily seen by using the explain api:
GET kibana_sample_data_ecommerce/_explain/ER5Bv3ABEiTwEf3FhKws
{
"query": {
"function_score": {
"functions": [
{
"filter": { "match": { "customer_last_name": "Cook" } },
"weight": 2
},
{
"filter": { "match": { "customer_first_name": "Jackson" } },
"weight": 4
},
{
"filter": { "match": { "customer_gender" : "MALE"} },
"weight": 8
}
],
"score_mode": "avg",
"boost_mode": "replace"
}
}
}
returns
{
"_index" : "kibana_sample_data_ecommerce",
"_type" : "_doc",
"_id" : "ER5Bv3ABEiTwEf3FhKws",
"matched" : true,
"explanation" : {
"value" : 1.0,
"description" : "min of:",
"details" : [
{
"value" : 1.0,
"description" : "function score, score mode [avg]",
"details" : [
{
"value" : 8.0,
"description" : "function score, product of:",
"details" : [
{
"value" : 1.0,
"description" : "match filter: customer_gender:MALE",
"details" : [ ]
},
{
"value" : 8.0,
"description" : "product of:",
"details" : [
{
"value" : 1.0,
"description" : "constant score 1.0 - no function provided",
"details" : [ ]
},
{
"value" : 8.0,
"description" : "weight",
"details" : [ ]
}
]
}
]
}
]
},
{
"value" : 3.4028235E38,
"description" : "maxBoost",
"details" : [ ]
}
]
}
}

Answer 2

It's already answered here. Since you used boost_mode:replace, only the function scores are used & the query score gets ignored.
Based on that, since your weights are the same, they "cancel each other out" to result in 1.