How to bulk load to oracle table when column has missing data NaN

[Newbee25]

I'm trying to bulk load a dataframe into an oracle table. The following error occurs:

_oracledb.exceptions.NotSupportedError: DPY-3013: unsupported Python type str for database type DB_TYPENUMBER

No error occurs when the records are inserted one by one.

Oracle table:

Column_Name	Data_Type
Make	VARCHAR2(20 BYTE)
Model	VARCHAR2(20 BYTE)
Weight	NUMBER
Year	NUMBER(38,0)

Python Code

import pandas as pd
import numpy as np

data = {
  'Make': ['Honda', 'Toyota', 'BMW', 'Honda'],
  'Model': ['Civic', 'Camry', 'X5', 'Accord'],
  'Weight': [1020.25, np.NaN, 1235.37, 940.9],
  'Year': [2021, 2023, 2019, np.NaN]
}

df = pd.DataFrame(data)

print(df)
print(df.dtypes)

try:
    with db_connection.cursor() as cursor:

        cursor.executemany('insert into myTable values(:1, :2, :3, :4)', df.fillna('').values.tolist()) 

    db_connection.commit()

except Exception as error:
     print(error) 

Please note the floats NaN must be inserted as null values to the table.

I attempted df['Weight'].fillna(float('NaN'), inplace=True)and df['Year'].fillna(float('NaN'), inplace=True) I also attempted to replace NaN with None.

It seems it is not recognizing None and NaN as valid types.
Is issue a possible bug?

I'm using the thick client mode and it is calling init_oracle_client()

Version

python-oracledb 1.4.1 Python 3.9.16

[anthony-tuininga]

You can use an input type handler to transform the np.NaN values to whatever value you choose. Or you can use the method you suggested. Both should work fine. The value None is definitely recognized as a null value. The value NaN is also supported -- but only for insert into a binary_double or binary_float column. NUMBER columns do not support those values. Can you provide the full code that failed when trying with a None value? If you still see a problem I'll give it a try.

[anthony-tuininga]

This has been confirmed to work:

import math

def input_type_handler(cursor, value, arraysize):
    if isinstance(value, float):
        return cursor.var(oracledb.DB_TYPE_NUMBER, arraysize=arraysize,
                          inconverter=lambda x: None if math.isnan(x) else x)
db_connection.inputtypehandler = input_type_handler

[cjbj]

Closing - no activity.