using sqlalchemy to load csv file into a database

Because of the power of SQLAlchemy, I'm also using it on a project. It's power comes from the object-oriented way of "talking" to a database instead of hardcoding SQL statements that can be a pain to manage. Not to mention, it's also a lot faster.

To answer your question bluntly, yes! Storing data from a CSV into a database using SQLAlchemy is a piece of cake. Here's a full working example (I used SQLAlchemy 1.0.6 and Python 2.7.6):

from numpy import genfromtxtfrom time import timefrom datetime import datetimefrom sqlalchemy import Column, Integer, Float, Datefrom sqlalchemy.ext.declarative import declarative_basefrom sqlalchemy import create_enginefrom sqlalchemy.orm import sessionmakerdef Load_Data(file_name):    data = genfromtxt(file_name, delimiter=',', skip_header=1, converters={0: lambda s: str(s)})    return data.tolist()Base = declarative_base()class Price_History(Base):    #Tell SQLAlchemy what the table name is and if there's any table-specific arguments it should know about    __tablename__ = 'Price_History'    __table_args__ = {'sqlite_autoincrement': True}    #tell SQLAlchemy the name of column and its attributes:    id = Column(Integer, primary_key=True, nullable=False)     date = Column(Date)    opn = Column(Float)    hi = Column(Float)    lo = Column(Float)    close = Column(Float)    vol = Column(Float)if __name__ == "__main__":    t = time()    #Create the database    engine = create_engine('sqlite:///csv_test.db')    Base.metadata.create_all(engine)    #Create the session    session = sessionmaker()    session.configure(bind=engine)    s = session()    try:        file_name = "t.csv" #sample CSV file used:  http://www.google.com/finance/historical?q=NYSE%3AT&ei=W4ikVam8LYWjmAGjhoHACw&output=csv        data = Load_Data(file_name)         for i in data:            record = Price_History(**{                'date' : datetime.strptime(i[0], '%d-%b-%y').date(),                'opn' : i[1],                'hi' : i[2],                'lo' : i[3],                'close' : i[4],                'vol' : i[5]            })            s.add(record) #Add all the records        s.commit() #Attempt to commit all the records    except:        s.rollback() #Rollback the changes on error    finally:        s.close() #Close the connection    print "Time elapsed: " + str(time() - t) + " s." #0.091s

(Note: this is not necessarily the "best" way to do this, but I think this format is very readable for a beginner; it's also very fast: 0.091s for 251 records inserted!)

I think if you go through it line by line, you'll see what a breeze it is to use. Notice the lack of SQL statements -- hooray! I also took the liberty of using numpy to load the CSV contents in two lines, but it can be done without it if you like.

If you wanted to compare against the traditional way of doing it, here's a full-working example for reference:

import sqlite3import timefrom numpy import genfromtxtdef dict_factory(cursor, row):    d = {}    for idx, col in enumerate(cursor.description):        d[col[0]] = row[idx]    return ddef Create_DB(db):          #Create DB and format it as needed    with sqlite3.connect(db) as conn:        conn.row_factory = dict_factory        conn.text_factory = str        cursor = conn.cursor()        cursor.execute("CREATE TABLE [Price_History] ([id] INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL UNIQUE, [date] DATE, [opn] FLOAT, [hi] FLOAT, [lo] FLOAT, [close] FLOAT, [vol] INTEGER);")def Add_Record(db, data):    #Insert record into table    with sqlite3.connect(db) as conn:        conn.row_factory = dict_factory        conn.text_factory = str        cursor = conn.cursor()        cursor.execute("INSERT INTO Price_History({cols}) VALUES({vals});".format(cols = str(data.keys()).strip('[]'),                     vals=str([data[i] for i in data]).strip('[]')                    ))def Load_Data(file_name):    data = genfromtxt(file_name, delimiter=',', skiprows=1, converters={0: lambda s: str(s)})    return data.tolist()if __name__ == "__main__":    t = time.time()     db = 'csv_test_sql.db' #Database filename     file_name = "t.csv" #sample CSV file used:  http://www.google.com/finance/historical?q=NYSE%3AT&ei=W4ikVam8LYWjmAGjhoHACw&output=csv    data = Load_Data(file_name) #Get data from CSV    Create_DB(db) #Create DB    #For every record, format and insert to table    for i in data:        record = {                'date' : i[0],                'opn' : i[1],                'hi' : i[2],                'lo' : i[3],                'close' : i[4],                'vol' : i[5]            }        Add_Record(db, record)    print "Time elapsed: " + str(time.time() - t) + " s." #3.604s

(Note: even in the "old" way, this is by no means the best way to do this, but it's very readable and a "1-to-1" translation from the SQLAlchemy way vs. the "old" way.)

Notice the the SQL statements: one to create the table, the other to insert records. Also, notice that it's a bit more cumbersome to maintain long SQL strings vs. a simple class attribute addition. Liking SQLAlchemy so far?

As for your foreign key inquiry, of course. SQLAlchemy has the power to do this too. Here's an example of how a class attribute would look like with a foreign key assignment (assuming the ForeignKey class has also been imported from the sqlalchemy module):

class Asset_Analysis(Base):    #Tell SQLAlchemy what the table name is and if there's any table-specific arguments it should know about    __tablename__ = 'Asset_Analysis'    __table_args__ = {'sqlite_autoincrement': True}    #tell SQLAlchemy the name of column and its attributes:    id = Column(Integer, primary_key=True, nullable=False)     fid = Column(Integer, ForeignKey('Price_History.id'))

which points the "fid" column as a foreign key to Price_History's id column.

Hope that helps!

In case your CSV is quite large, using INSERTS is very ineffective. You should use a bulk loading mechanisms, which differ from base to base. E.g. in PostgreSQL you should use "COPY FROM" method:

with open(csv_file_path, 'r') as f:        conn = create_engine('postgresql+psycopg2://...').raw_connection()    cursor = conn.cursor()    cmd = 'COPY tbl_name(col1, col2, col3) FROM STDIN WITH (FORMAT CSV, HEADER FALSE)'    cursor.copy_expert(cmd, f)    conn.commit()

I have had the exact same problem, and I found it paradoxically easier to use a 2-step process with pandas:

import pandas as pdwith open(csv_file_path, 'r') as file:    data_df = pd.read_csv(file)data_df.to_sql('tbl_name', con=engine, index=True, index_label='id', if_exists='replace')

Note that my approach is similar to this one, but somehow Google sent me to this thread instead, so I thought I would share.