mongodb avg in aggregation of array element

1. The $avg group accumulator operator does work, it's only that in your case it is being applied to an element in an array and thus gives the "incorrect" result.

2. When you use the $max group accumulator operator, it returns the the highest value that results from applying an expression to each document in a group of documents, thus in your example it returned the maximum array.

To demonstrate this, consider adding a few sample documents to a test collection in mongoshell:

db.test.insert([
{
    "_id" : {
        "d_timestamp" : NumberLong(1429949699),
        "d_isostamp" : ISODate("2015-04-25T08:14:59.000Z")
    },
    "XBT-USD-stp-nex" : [ 
        {
            "sellSpread" : -1.80081
        }
    ]
},
{
    "_id" : {
        "d_timestamp" : NumberLong(1429949710),
        "d_isostamp" : ISODate("2015-04-25T08:15:10.000Z")
    },
    "XBT-USD-stp-nex" : [ 
        {
            "sellSpread" : -1.80079
        }
    ]
},
{
    "_id" : {
        "d_timestamp" : NumberLong(1429949720),
        "d_isostamp" : ISODate("2015-04-25T08:15:20.000Z")
    },
    "XBT-USD-stp-nex" : [ 
        {
            "sellSpread" : -1.80083
        }
    ]
},
{
    "_id" : {
        "d_timestamp" : NumberLong(1429949730),
        "d_isostamp" : ISODate("2015-04-25T08:15:30.000Z")
    },
    "XBT-USD-stp-nex" : [ 
        {
            "sellSpread" : -1.80087
        }
    ]
}
])

Now, replicating the same operation above in mongoshell:

var spreadName = "XBT-USD-stp-nex",
    start = new Date(2015, 3, 25),
    end = new Date(2015, 3, 26);
db.test.aggregate([
    {
        "$match": {
            "_id.d_isostamp": { "$gte": start, "$lte": end }
        }
    },
    {
        "$project": {
            "sellSpread": "$"+spreadName+".sellSpread"
        }
    }/*,<--- deliberately omitted the $unwind stage from the pipeline to replicate the current pipeline
    {
        "$unwind": "$sellSpread"
    }*/,
    {
        "$group": {
            "_id": {
                "isodate": { "$minute": "$_id.d_isostamp"}
            },
            "rsell_spread": {
                "$avg": "$sellSpread"
            }
        }
    }
])

Output:

/* 0 */
{
    "result" : [ 
        {
            "_id" : {
                "isodate" : 15
            },
            "rsell_spread" : 0
        }, 
        {
            "_id" : {
                "isodate" : 14
            },
            "rsell_spread" : 0
        }
    ],
    "ok" : 1
}

The solution is to include an $unwind operator pipeline stage after the $project step, this will deconstruct the XBT-USD-stp-nex array field from the input documents and outputs a document for each element. Each output document replaces the array with an element value. This will then make it possible for the $avg group accumulator operator to work.

Including this will give the aggregation result:

/* 0 */
{
    "result" : [ 
        {
            "_id" : {
                "isodate" : 15
            },
            "rsell_spread" : -1.80083
        }, 
        {
            "_id" : {
                "isodate" : 14
            },
            "rsell_spread" : -1.80081
        }
    ],
    "ok" : 1
}

So your final working aggregation in PHP should be:

$spreadName ='XBT-USD-stp-nex';
$pipe = array(
    array(
        '$match' => array(
            '_id.d_isostamp' => array(
                '$gt' => $start, '$lt' => $end
            )
        )
    ),    
    array(
        '$project' => array(
            'sellSpread' =>'$'.$spreadName.'.sellSpread',
        )
    ),
    array('$unwind' => '$sellSpread'),
    array(
        '$group' => array(
            '_id' => array(
                'isodate' => array(
                    '$minute' => '$_id.d_isostamp'
                )
            ),
            'rsell_spread' => array(
                '$avg' => '$sellSpread'
            ),
        )
    ),
);

$out = $collection->aggregate($pipe ,$options);