SQLAlchemy ORM conversion to pandas DataFrame

If you want to compile a query with parameters and dialect specific arguments, use something like this:

c = query.statement.compile(query.session.bind)
df = pandas.read_sql(c.string, query.session.bind, params=c.params)

Just to make this more clear for novice pandas programmers, here is a concrete example,

pd.read_sql(session.query(Complaint).filter(Complaint.id == 2).statement,session.bind) 

Here we select a complaint from complaints table (sqlalchemy model is Complaint) with id = 2

The selected solution didn’t work for me, as I kept getting the error

AttributeError: ‘AnnotatedSelect’ object has no attribute ‘lower’

I found the following worked:

df = pd.read_sql_query(query.statement, engine)

from sqlalchemy import Column, Integer, String, create_engine
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy.orm import sessionmaker

engine = create_engine('postgresql://postgres:postgres@localhost:5432/DB', echo=False)
Base = declarative_base(bind=engine)
Session = sessionmaker(bind=engine)
session = Session()

conn = session.bind

class DailyTrendsTable(Base):

    __tablename__ = 'trends'
    __table_args__ = ({"schema": 'mf_analysis'})

    company_code = Column(DOUBLE_PRECISION, primary_key=True)
    rt_bullish_trending = Column(Integer)
    rt_bearish_trending = Column(Integer)
    rt_bullish_non_trending = Column(Integer)
    rt_bearish_non_trending = Column(Integer)
    gen_date = Column(Date, primary_key=True)

df_query = select([DailyTrendsTable])

df_data = pd.read_sql(rt_daily_query, con = conn)

For completeness sake: As alternative to the Pandas-function read_sql_query(), you can also use the Pandas-DataFrame-function from_records() to convert a structured or record ndarray to DataFrame.
This comes in handy if you e.g. have already executed the query in SQLAlchemy and have the results already available:

import pandas as pd 
from sqlalchemy import Column, Integer, String, create_engine
from sqlalchemy.ext.declarative import declarative_base
from sqlalchemy.orm import scoped_session, sessionmaker


SQLALCHEMY_DATABASE_URI = 'postgresql://postgres:postgres@localhost:5432/my_database'
engine = create_engine(SQLALCHEMY_DATABASE_URI, pool_pre_ping=True, echo=False)
db = scoped_session(sessionmaker(autocommit=False, autoflush=False, bind=engine))
Base = declarative_base(bind=engine)


class Currency(Base):
    """The `Currency`-table"""
    __tablename__ = "currency"
    __table_args__ = {"schema": "data"}

    id = Column(Integer, primary_key=True, nullable=False)
    name = Column(String(64), nullable=False)


# Defining the SQLAlchemy-query
currency_query = db.query(Currency).with_entities(Currency.id, Currency.name)

# Getting all the entries via SQLAlchemy
currencies = currency_query.all()

# We provide also the (alternate) column names and set the index here,
# renaming the column `id` to `currency__id`
df_from_records = pd.DataFrame.from_records(currencies
    , index='currency__id'
    , columns=['currency__id', 'name'])
print(df_from_records.head(5))

# Or getting the entries via Pandas instead of SQLAlchemy using the
# aforementioned function `read_sql_query()`. We can set the index-columns here as well
df_from_query = pd.read_sql_query(currency_query.statement, db.bind, index_col='id')
# Renaming the index-column(s) from `id` to `currency__id` needs another statement
df_from_query.index.rename(name='currency__id', inplace=True)
print(df_from_query.head(5))

This answer provides a reproducible example using an SQL Alchemy select statement and returning a pandas data frame. It is based on an in memory SQLite database so that anyone can reproduce it without installing a database engine.

import pandas
from sqlalchemy import create_engine
from sqlalchemy import MetaData, Table, Column, Text
from sqlalchemy.orm import Session

Define table metadata and create a table

engine = create_engine('sqlite://')
meta = MetaData()
meta.bind = engine
user_table = Table('user', meta,
                   Column("name", Text),
                   Column("full_name", Text))
user_table.create()

Insert some data into the user table

stmt = user_table.insert().values(name='Bob', full_name='Sponge Bob')
with Session(engine) as session:
    result = session.execute(stmt)
    session.commit()

Read the result of a select statement into a pandas data frame

# Select data into a pandas data frame
stmt = user_table.select().where(user_table.c.name == 'Bob')
df = pandas.read_sql_query(stmt, engine)
df
Out:
  name   full_name
0  Bob  Sponge Bob

if use SQL query

def generate_df_from_sqlquery(query):
   from pandas import DataFrame
   query = db.session.execute(query)
   df = DataFrame(query.fetchall())
   if len(df) > 0:
      df.columns = query.keys()
   else:
      columns = query.keys()
      df = pd.DataFrame(columns=columns)
return df

profile_df = generate_df_from_sqlquery(profile_query) 

Using the 2.0 SQLalchemy syntax (available also in 1.4 with the flag future=True) it looks that pd.read_sql is not implemented yet and it will raise:

NotImplementedError: This method is not implemented for SQLAlchemy 2.0.

This is an open issue that won’t be solved till pandas 2.0, you can find some information about this here and here.

I didn’t find any satisfactory work around, but some people seems to be using two configurations of the engine, one with the flag future False:

engine2 = create_engine(URL_string, echo=False, future=False)

This solution would be OK if you query strings, but using the ORM, the best I could do is a custom function yet to be optimized, but it works:

Conditions = session.query(ExampleTable)
def df_from_sql(query):
    return pd.DataFrame([i.__dict__ for i in query]).drop(columns='_sa_instance_state')
df = df_from_sql(ExampleTable)

This solution in any case would be provisional till pd.read_sql has implemented the new syntax.

When you’re using the ORM it’s as simple as this:

pd.DataFrame([r._asdict() for r in query.all()])

Good alternative to pd.read_sql when you don’t want to expose sql and sessions to the business logic code.

Found it here: https://stackoverflow.com/a/52208023/1635525

Simple example using the Result.keys() method to get the column names.

import sqlalchemy as sa
import pandas as pd

engine = sa.create_engine(...)
with engine.connect() as conn:
   result = conn.execute("SELECT * FROM foo;")
   df = pd.DataFrame(result.all(), columns=result.keys())

https://docs.sqlalchemy.org/en/20/core/connections.html#sqlalchemy.engine.Result.keys

Adding to answers using read_sql like @van, when my query involved a join, sqlalchemy seemed to be implicitly adding aliased columns from the join tables like id_1, id_2 incase the join tables and primary table both had an id column for example. Using .all() removes these implicit columns before returning results but read_sql will include these columns.

Solutions for that case for me was to be explicit on my selects. So I replaced

query = session.query(model)

with

query = session.query(model.col_1, model.col_2)

or for select all

query = session.query(*model.__table__.columns.values())

then

df = pd.read_sql(query.statement, query.session.bind)

Another way to do this if you’ve already received the data after sqalchemy filtering is presented here
Followed by this answer, use this function:

from collections import defaultdict
from sqlalchemy import inspect
import pandas as pd
def query_to_dict(rest):
    result = defaultdict(list)
    for obj in rest:
        instance = inspect(obj)
        for key, x in instance.attrs.items():
            result[key].append(x.value)
    return result
# Getting all the entries via SQLAlchemy
currencies = currency_query.all()
df = pd.DataFrame(query_to_dict(rset))

for me this worked for me in 2023: First create a engine and session, convert the query-string into a SQLalchemy text

from sqlalchemy import text

    engine = create_engine(path, echo=True)

    Session = sessionmaker(bind=engine)
    session = Session()

    # generates suitble text for SQLalchemy
    sql = text(query_string)

    session.execute(sql)

    df = pd.read_sql_query(sql=sql, con=engine.connect())