Merge CSVs in Python with different columns
Question:
I have hundreds of large CSV files that I would like to merge into one. However, not all CSV files contain all columns. Therefore, I need to merge files based on column name, not column position.
Just to be clear: in the merged CSV, values should be empty for a cell coming from a line which did not have the column of that cell.
I cannot use the pandas module, because it makes me run out of memory.
Is there a module that can do that, or some easy code?
Answers:
The csv.DictReader and csv.DictWriter classes should work well (see Python docs). Something like this:
import csv
inputs = ["in1.csv", "in2.csv"] # etc
# First determine the field names from the top line of each input file
# Comment 1 below
fieldnames = []
for filename in inputs:
with open(filename, "r", newline="") as f_in:
reader = csv.reader(f_in)
headers = next(reader)
for h in headers:
if h not in fieldnames:
fieldnames.append(h)
# Then copy the data
with open("out.csv", "w", newline="") as f_out: # Comment 2 below
writer = csv.DictWriter(f_out, fieldnames=fieldnames)
for filename in inputs:
with open(filename, "r", newline="") as f_in:
reader = csv.DictReader(f_in) # Uses the field names in this file
for line in reader:
# Comment 3 below
writer.writerow(line)
Comments from above:
- You need to specify all the possible field names in advance to
DictWriter, so you need to loop through all your CSV files twice: once to find all the headers, and once to read the data. There is no better solution, because all the headers need to be known before DictWriter can write the first line. This part would be more efficient using sets instead of lists (the in operator on a list is comparatively slow), but it won’t make much difference for a few hundred headers. Sets would also lose the deterministic ordering of a list – your columns would come out in a different order each time you ran the code.
- The above code is for Python 3, where weird things happen in the CSV module without
newline="". Remove this for Python 2.
- At this point,
line is a dict with the field names as keys, and the column data as values. You can specify what to do with blank or unknown values in the DictReader and DictWriter constructors.
This method should not run out of memory, because it never has the whole file loaded at once.
For those of us using 2.7, this adds an extra linefeed between records in “out.csv”. To resolve this, just change the file mode from “w” to “wb”.
You can use the pandas module to do this pretty easily. This snippet assumes all your csv files are in the current folder.
import pandas as pd
import os
all_csv = [file_name for file_name in os.listdir(os.getcwd()) if '.csv' in file_name]
li = []
for filename in all_csv:
df = pd.read_csv(filename, index_col=None, header=0, parse_dates=True, infer_datetime_format=True)
li.append(df)
frame = pd.concat(li, axis=0, ignore_index=True)
frame.to_csv('melted_csv.csv', index=False)
The solution by @Aaron Lockey, which is the accepted answer has worked well for me except, there were no headers for the file. The out put had no headers and only the row data. Each column was without headings (keys). So I inserted following:
writer.writeheader()
and it worked perfectly fine for me! So now the entire code appears like this:
import csv
inputs = ["in1.csv", "in2.csv"] # etc
# First determine the field names from the top line of each input file
fieldnames = []
for filename in inputs:
with open(filename, "r", newline="") as f_in:
reader = csv.reader(f_in)
headers = next(reader)
for h in headers:
if h not in fieldnames:
fieldnames.append(h)
# Then copy the data
with open("out.csv", "w", newline="") as f_out:
writer = csv.DictWriter(f_out, fieldnames=fieldnames)
writer.writeheader() #this is the addition.
for filename in inputs:
with open(filename, "r", newline="") as f_in:
reader = csv.DictReader(f_in) # Uses the field names in this file
for line in reader:
writer.writerow(line)
I’ve faced a situation where not only the number of columns are different, but also some column names are missing. For this kind of situation and obviously for your case, this code snippet can help you 🙂
The tricky part is naming the columns which have no names and adding them to dictionary. The read_csv_file function is playing the main role here.
def read_csv_file(csv_file_path):
headers = []
data = []
with open(csv_file_path, 'r') as f:
csv_reader = csv.reader(f)
rows = []
for i, row in enumerate(csv_reader):
if i == 0:
for j in range(len(row)):
if row[j].strip() == "":
col_name = f"col-{j+1}"
else:
col_name = row[j]
if col_name not in headers:
headers.append(col_name)
else:
rows.append(row)
if len(row) > len(headers):
for j in range(len(row)):
if j+1 > len(headers):
col_name = f"col-{j+1}"
if col_name not in headers:
headers.append(col_name)
for i, row in enumerate(rows):
row_data = {}
for j in range(len(headers)):
if len(row) > j:
row_data[headers[j]] = row[j]
else:
row_data[headers[j]] = ''
data.append(row_data)
return headers, data
def write_csv_file(file_path, rows):
if len(rows) > 0:
headers = list(rows[0].keys())
with open(file_path, 'w', newline='', encoding='UTF8') as f:
writer = csv.DictWriter(f, fieldnames=headers)
writer.writeheader()
writer.writerows(rows)
# The list of the csv file paths which will be merged
files_to_be_merged = [
'file-1.csv',
'file-2.csv',
'file-3.csv'
]
# Read and store all the file data in new_file_data
final_headers = []
new_file_data = []
for f1 in files_to_be_merged:
single_file_data = read_csv_file(f1)
for h in single_file_data[0]:
if h not in final_headers:
final_headers.append(h)
new_file_data += single_file_data[1]
# Add the missing keys to the dictionaries
for d in new_file_data:
for h in final_headers:
if d.get(h) is None:
d[h] = ""
# Write a new file
target_file_name = 'merged_file.csv'
write_csv_file(target_file_name, new_file_data)
I have hundreds of large CSV files that I would like to merge into one. However, not all CSV files contain all columns. Therefore, I need to merge files based on column name, not column position.
Just to be clear: in the merged CSV, values should be empty for a cell coming from a line which did not have the column of that cell.
I cannot use the pandas module, because it makes me run out of memory.
Is there a module that can do that, or some easy code?
The csv.DictReader and csv.DictWriter classes should work well (see Python docs). Something like this:
import csv
inputs = ["in1.csv", "in2.csv"] # etc
# First determine the field names from the top line of each input file
# Comment 1 below
fieldnames = []
for filename in inputs:
with open(filename, "r", newline="") as f_in:
reader = csv.reader(f_in)
headers = next(reader)
for h in headers:
if h not in fieldnames:
fieldnames.append(h)
# Then copy the data
with open("out.csv", "w", newline="") as f_out: # Comment 2 below
writer = csv.DictWriter(f_out, fieldnames=fieldnames)
for filename in inputs:
with open(filename, "r", newline="") as f_in:
reader = csv.DictReader(f_in) # Uses the field names in this file
for line in reader:
# Comment 3 below
writer.writerow(line)
Comments from above:
- You need to specify all the possible field names in advance to
DictWriter, so you need to loop through all your CSV files twice: once to find all the headers, and once to read the data. There is no better solution, because all the headers need to be known beforeDictWritercan write the first line. This part would be more efficient using sets instead of lists (theinoperator on a list is comparatively slow), but it won’t make much difference for a few hundred headers. Sets would also lose the deterministic ordering of a list – your columns would come out in a different order each time you ran the code. - The above code is for Python 3, where weird things happen in the CSV module without
newline="". Remove this for Python 2. - At this point,
lineis a dict with the field names as keys, and the column data as values. You can specify what to do with blank or unknown values in theDictReaderandDictWriterconstructors.
This method should not run out of memory, because it never has the whole file loaded at once.
For those of us using 2.7, this adds an extra linefeed between records in “out.csv”. To resolve this, just change the file mode from “w” to “wb”.
You can use the pandas module to do this pretty easily. This snippet assumes all your csv files are in the current folder.
import pandas as pd
import os
all_csv = [file_name for file_name in os.listdir(os.getcwd()) if '.csv' in file_name]
li = []
for filename in all_csv:
df = pd.read_csv(filename, index_col=None, header=0, parse_dates=True, infer_datetime_format=True)
li.append(df)
frame = pd.concat(li, axis=0, ignore_index=True)
frame.to_csv('melted_csv.csv', index=False)
The solution by @Aaron Lockey, which is the accepted answer has worked well for me except, there were no headers for the file. The out put had no headers and only the row data. Each column was without headings (keys). So I inserted following:
writer.writeheader()
and it worked perfectly fine for me! So now the entire code appears like this:
import csv
inputs = ["in1.csv", "in2.csv"] # etc
# First determine the field names from the top line of each input file
fieldnames = []
for filename in inputs:
with open(filename, "r", newline="") as f_in:
reader = csv.reader(f_in)
headers = next(reader)
for h in headers:
if h not in fieldnames:
fieldnames.append(h)
# Then copy the data
with open("out.csv", "w", newline="") as f_out:
writer = csv.DictWriter(f_out, fieldnames=fieldnames)
writer.writeheader() #this is the addition.
for filename in inputs:
with open(filename, "r", newline="") as f_in:
reader = csv.DictReader(f_in) # Uses the field names in this file
for line in reader:
writer.writerow(line)
I’ve faced a situation where not only the number of columns are different, but also some column names are missing. For this kind of situation and obviously for your case, this code snippet can help you 🙂
The tricky part is naming the columns which have no names and adding them to dictionary. The read_csv_file function is playing the main role here.
def read_csv_file(csv_file_path):
headers = []
data = []
with open(csv_file_path, 'r') as f:
csv_reader = csv.reader(f)
rows = []
for i, row in enumerate(csv_reader):
if i == 0:
for j in range(len(row)):
if row[j].strip() == "":
col_name = f"col-{j+1}"
else:
col_name = row[j]
if col_name not in headers:
headers.append(col_name)
else:
rows.append(row)
if len(row) > len(headers):
for j in range(len(row)):
if j+1 > len(headers):
col_name = f"col-{j+1}"
if col_name not in headers:
headers.append(col_name)
for i, row in enumerate(rows):
row_data = {}
for j in range(len(headers)):
if len(row) > j:
row_data[headers[j]] = row[j]
else:
row_data[headers[j]] = ''
data.append(row_data)
return headers, data
def write_csv_file(file_path, rows):
if len(rows) > 0:
headers = list(rows[0].keys())
with open(file_path, 'w', newline='', encoding='UTF8') as f:
writer = csv.DictWriter(f, fieldnames=headers)
writer.writeheader()
writer.writerows(rows)
# The list of the csv file paths which will be merged
files_to_be_merged = [
'file-1.csv',
'file-2.csv',
'file-3.csv'
]
# Read and store all the file data in new_file_data
final_headers = []
new_file_data = []
for f1 in files_to_be_merged:
single_file_data = read_csv_file(f1)
for h in single_file_data[0]:
if h not in final_headers:
final_headers.append(h)
new_file_data += single_file_data[1]
# Add the missing keys to the dictionaries
for d in new_file_data:
for h in final_headers:
if d.get(h) is None:
d[h] = ""
# Write a new file
target_file_name = 'merged_file.csv'
write_csv_file(target_file_name, new_file_data)