PostgreSQL / Pandas插入包含null / np.nan / empty值的稀疏数组

I want to insert an array that contains empty values created from Pandas in Python, and these empty values default to np.nan in Pandas dataframe. I don't want them to be 'NaN' in my PostgreSQL database, I want my PostgreSQL arrays to contain empty values like this: '{123,24,,23}' so they are not counted in my aggregate functions like calculating the mean or standard deviation across indices. I am not sure if it is possible to have sparse arrays in PostgreSQL. There won't be a lot of sparse arrays in my dataset, I am just testing this for edge case purposes.

我想插入一个包含在Python中从Pandas创建的空值的数组,这些空值默认为Pandas数据帧中的np.nan。我不希望它们在我的PostgreSQL数据库中成为'NaN',我希望我的PostgreSQL数组包含这样的空值:'{123,24,,23}'因此它们不计入我的聚合函数中,如计算指数的均值或标准差。我不确定PostgreSQL中是否有可能有稀疏数组。我的数据集中不会有很多稀疏数组,我只是为了边缘情况测试它。

My table schema:

我的表架构:

create_table = '''
            CREATE TABLE {t} (
                patient_id VARCHAR[20] PRIMARY KEY,
                gene_expression double precision []
            );
        '''

The relevant Python code (I don't know how to write the proper SQL code here). Here I converted the array into a string, because Python arrays cannot be sparse:

相关的Python代码(我不知道如何在这里编写正确的SQL代码)。在这里,我将数组转换为字符串,因为Python数组不能稀疏:

df = df.fillna('')
NCI = [1]
MCI = [2,3]
AD = [4,5]
other = [6]

insert_sql = '''
                INSERT INTO {t} (patient_id, gene_expression)
                VALUES (%s,%s);
            '''
cur = psql_conn.cursor()

for index, row in df.iterrows():
    arr = row[2:].tolist()
    postgres_arr = ','.join(map(str, arr))
    if row['DIAGNOSIS'].isdigit():
        if int(row['DIAGNOSIS']) in NCI:
            cur.execute(insert_sql.format(t='nci'), (row['PATIENT_ID'], postgres_arr,))

        elif int(row['DIAGNOSIS']) in MCI:
            cur.execute(insert_sql.format(t='mci'), (row['PATIENT_ID'], postgres_arr,))

        elif int(row['DIAGNOSIS']) in AD:
            cur.execute(insert_sql.format(t='ad'), (row['PATIENT_ID'], postgres_arr,))

        elif int(row['DIAGNOSIS']) in other:
            cur.execute(insert_sql.format(t='other'), (row['PATIENT_ID'], postgres_arr,))

    elif row['DIAGNOSIS'] == '':
        cur.execute(insert_sql.format(t='na'), (row['PATIENT_ID'], postgres_arr,))

    else:
        print('ERROR: unknown diagnosis {d}.'.format(d=diagnosis))

psql_conn.commit()
cur.close()

My Error:

psycopg2.DataError: malformed array literal: "{2.0,2.4,}"
LINE 3:                     VALUES ('X100_120417','{2.0,2.4,}');
                                                  ^
DETAIL:  Unexpected "}" character.

2 个解决方案

#1

If you want to create a column with max length, use bracket, not square brackets. change VARCHAR[20] to VARCHAR(20) in create table statement. Otherwise first %s is expected to be array and it is varchar. Here is sample - mind that patient_id is created as array, not varchar...

如果要创建具有最大长度的列,请使用括号,而不是方括号。在create table语句中将VARCHAR [20]更改为VARCHAR(20)。否则,首先%s应该是数组,它是varchar。这是示例 - 请注意,patient_id是作为数组创建的,而不是varchar ...

t=# CREATE TABLE so23 (
                patient_id VARCHAR[20] PRIMARY KEY,
                gene_expression double precision []
            );
CREATE TABLE
t=# \d+ so23
                                    Table "public.so23"
     Column      |        Type         | Modifiers | Storage  | Stats target | Description
-----------------+---------------------+-----------+----------+--------------+-------------
 patient_id      | character varying[] | not null  | extended |              |
 gene_expression | double precision[]  |           | extended |              |
Indexes:
    "so23_pkey" PRIMARY KEY, btree (patient_id)

#2

After a few hours of trial and error:

经过几个小时的反复试验:

Load this Pandas dataframe df from some CSV file:

从某些CSV文件加载此Pandas数据帧df:

+----+-------+--------------+
| id | stuff |    array     |
+----+-------+--------------+
|  0 | a     | {1,2,3}      |
|  1 | b     | {1,np.nan,3} |
|  2 | 45    | {np.nan,4,2} |
+----+-------+--------------+

process in pandas using:

大熊猫进程使用:

df = df.fillna('NULL')
insert_sql = '''
                INSERT INTO {t} (patient_id, gene_expression)
                VALUES (%s,%s);
            '''

for index, row in df.iterrows():
    arr = row[2:].tolist()
    postgres_arr = '{' + ','.join(map(str,arr)) + '}'
    cur.execute(insert_sql.format(t='my_table'), (row['id'], postgres_arr,))

My main issue was recognizing that string literal 'NULL' automatically translate to PostgreSQL NULL keyword, which is ignored in calculations and results of aggregate functions return a value as if the NULL values are not there, versus NaN keyword where every operation with it results in NaN.

我的主要问题是识别字符串文字'NULL'自动转换为PostgreSQL NULL关键字,在计算中忽略它,并且聚合函数的结果返回一个值,好像NULL值不存在,而NaN关键字,其中每个操作都会导致NaN的。