Hive SerDe initialize() function called 8 times - hadoop

Question

I'm writing a custom Hive SerDe that can read data stored as ORC.
The underlying table contains an event_map column MAP. The SerDe reads data from the event_map column and explodes them into separate columns after some additional processing. Eg: The map would be {'key1': 'val1', 'key2': 'val2'}.
Then i'll create an external table like:
ADD JAR hdfs:///user/my-serde.jar;
CREATE EXTERNAL TABLE IF NOT EXISTS dev_db.expand_data_using_serde (
key1 STRING,
key2 STRING
) PARTITIONED BY (
ds STRING
)
ROW FORMAT SERDE 'com.....MySerDeClass'
WITH SERDEPROPERTIES (
"key1" = "extract_json",
"key2" = "multiply_by_5"
)
STORED AS ORC
LOCATION 'hdfs://dev.db/underlying_table_stored_as_orc';
USE dev_db;
MSCK REPAIR TABLE expand_data_using_serde;
Then the SerDe would find the 'key1' and 'key2' columns and extract it from the underlying map and expose them as separate columns instead of a MAP. I know we can explode this column through a hive query but providing this example for simplicity. i.e. The SerDe would check if the table DDL has key1 as a column, and would extract key1 from the event_map and use it for that column.
Majority of the processing is in the deserialize() function of the serde. But, i notice that my initialize() method is being called 8 times.
Why is that? The deserialize is called once per row which makes sense but the initialize is called repeatedly.
Also, the SerDe has access to the output table schema through the "columns" property, can it access the query as well?
Eg: If my query is
SELECT key1 FROM dev_db.expand_data_using_serde;
Can the serde know that we only need to read key1 and skip the processing for key2?