I would like to read several CSV files from a directory into pandas and concatenate them into one big DataFrame. I have not been able to figure it out though. Here is what I have so far:
import glob
import pandas as pd
# Get data file names
path = r'C:\DRO\DCL_rawdata_files'
filenames = glob.glob(path + "/*.csv")
dfs = []
for filename in filenames:
dfs.append(pd.read_csv(filename))
# Concatenate all data into one DataFrame
big_frame = pd.concat(dfs, ignore_index=True)
I guess I need some help within the for loop?
See pandas: IO tools for all of the available .read_ methods.
Try the following code if all of the CSV files have the same columns.
I have added header=0, so that after reading the CSV file's first row, it can be assigned as the column names.
import pandas as pd
import glob
import os
path = r'C:\DRO\DCL_rawdata_files' # use your path
all_files = glob.glob(os.path.join(path , "/*.csv"))
li = []
for filename in all_files:
df = pd.read_csv(filename, index_col=None, header=0)
li.append(df)
frame = pd.concat(li, axis=0, ignore_index=True)
Or, with attribution to a comment from Sid.
all_files = glob.glob(os.path.join(path, "*.csv"))
df = pd.concat((pd.read_csv(f) for f in all_files), ignore_index=True)
It's often necessary to identify each sample of data, which can be accomplished by adding a new column to the dataframe.
pathlib from the standard library will be used for this example. It treats paths as objects with methods, instead of strings to be sliced.
Imports and Setup
from pathlib import Path
import pandas as pd
import numpy as np
path = r'C:\DRO\DCL_rawdata_files' # or unix / linux / mac path
# Get the files from the path provided in the OP
files = Path(path).glob('*.csv') # .rglob to get subdirectories
Option 1:
Add a new column with the file name
dfs = list()
for f in files:
data = pd.read_csv(f)
# .stem is method for pathlib objects to get the filename w/o the extension
data['file'] = f.stem
dfs.append(data)
df = pd.concat(dfs, ignore_index=True)
Option 2:
Add a new column with a generic name using enumerate
dfs = list()
for i, f in enumerate(files):
data = pd.read_csv(f)
data['file'] = f'File {i}'
dfs.append(data)
df = pd.concat(dfs, ignore_index=True)
Option 3:
Create the dataframes with a list comprehension, and then use np.repeat to add a new column.
[f'S{i}' for i in range(len(dfs))] creates a list of strings to name each dataframe.
[len(df) for df in dfs] creates a list of lengths
Attribution for this option goes to this plotting answer.
# Read the files into dataframes
dfs = [pd.read_csv(f) for f in files]
# Combine the list of dataframes
df = pd.concat(dfs, ignore_index=True)
# Add a new column
df['Source'] = np.repeat([f'S{i}' for i in range(len(dfs))], [len(df) for df in dfs])
Option 4:
One liners using .assign to create the new column, with attribution to a comment from C8H10N4O2
df = pd.concat((pd.read_csv(f).assign(filename=f.stem) for f in files), ignore_index=True)
or
df = pd.concat((pd.read_csv(f).assign(Source=f'S{i}') for i, f in enumerate(files)), ignore_index=True)
An alternative to darindaCoder's answer:
path = r'C:\DRO\DCL_rawdata_files' # use your path
all_files = glob.glob(os.path.join(path, "*.csv")) # advisable to use os.path.join as this makes concatenation OS independent
df_from_each_file = (pd.read_csv(f) for f in all_files)
concatenated_df = pd.concat(df_from_each_file, ignore_index=True)
# doesn't create a list, nor does it append to one
import glob
import os
import pandas as pd
df = pd.concat(map(pd.read_csv, glob.glob(os.path.join('', "my_files*.csv"))))
Almost all of the answers here are either unnecessarily complex (glob pattern matching) or rely on additional third-party libraries. You can do this in two lines using everything Pandas and Python (all versions) already have built in.
For a few files - one-liner
df = pd.concat(map(pd.read_csv, ['d1.csv', 'd2.csv','d3.csv']))
For many files
import os
filepaths = [f for f in os.listdir(".") if f.endswith('.csv')]
df = pd.concat(map(pd.read_csv, filepaths))
For No Headers
If you have specific things you want to change with pd.read_csv (i.e., no headers) you can make a separate function and call that with your map:
def f(i):
return pd.read_csv(i, header=None)
df = pd.concat(map(f, filepaths))
This pandas line, which sets the df, utilizes three things:
Python's map (function, iterable) sends to the function (the
pd.read_csv()) the iterable (our list) which is every CSV element
in filepaths).
Panda's read_csv() function reads in each CSV file as normal.
Panda's concat() brings all these under one df variable.
Easy and Fast
Import two or more CSV files without having to make a list of names.
import glob
import pandas as pd
df = pd.concat(map(pd.read_csv, glob.glob('data/*.csv')))
The Dask library can read a dataframe from multiple files:
>>> import dask.dataframe as dd
>>> df = dd.read_csv('data*.csv')
(Source: https://examples.dask.org/dataframes/01-data-access.html#Read-CSV-files)
The Dask dataframes implement a subset of the Pandas dataframe API. If all the data fits into memory, you can call df.compute() to convert the dataframe into a Pandas dataframe.
I googled my way into Gaurav Singh's answer.
However, as of late, I am finding it faster to do any manipulation using NumPy and then assigning it once to a dataframe rather than manipulating the dataframe itself on an iterative basis and it seems to work in this solution too.
I do sincerely want anyone hitting this page to consider this approach, but I don't want to attach this huge piece of code as a comment and making it less readable.
You can leverage NumPy to really speed up the dataframe concatenation.
import os
import glob
import pandas as pd
import numpy as np
path = "my_dir_full_path"
allFiles = glob.glob(os.path.join(path,"*.csv"))
np_array_list = []
for file_ in allFiles:
df = pd.read_csv(file_,index_col=None, header=0)
np_array_list.append(df.as_matrix())
comb_np_array = np.vstack(np_array_list)
big_frame = pd.DataFrame(comb_np_array)
big_frame.columns = ["col1", "col2"....]
Timing statistics:
total files :192
avg lines per file :8492
--approach 1 without NumPy -- 8.248656988143921 seconds ---
total records old :1630571
--approach 2 with NumPy -- 2.289292573928833 seconds ---
A one-liner using map, but if you'd like to specify additional arguments, you could do:
import pandas as pd
import glob
import functools
df = pd.concat(map(functools.partial(pd.read_csv, sep='|', compression=None),
glob.glob("data/*.csv")))
Note: map by itself does not let you supply additional arguments.
If you want to search recursively (Python 3.5 or above), you can do the following:
from glob import iglob
import pandas as pd
path = r'C:\user\your\path\**\*.csv'
all_rec = iglob(path, recursive=True)
dataframes = (pd.read_csv(f) for f in all_rec)
big_dataframe = pd.concat(dataframes, ignore_index=True)
Note that the three last lines can be expressed in one single line:
df = pd.concat((pd.read_csv(f) for f in iglob(path, recursive=True)), ignore_index=True)
You can find the documentation of ** here. Also, I used iglobinstead of glob, as it returns an iterator instead of a list.
EDIT: Multiplatform recursive function:
You can wrap the above into a multiplatform function (Linux, Windows, Mac), so you can do:
df = read_df_rec('C:\user\your\path', *.csv)
Here is the function:
from glob import iglob
from os.path import join
import pandas as pd
def read_df_rec(path, fn_regex=r'*.csv'):
return pd.concat((pd.read_csv(f) for f in iglob(
join(path, '**', fn_regex), recursive=True)), ignore_index=True)
Inspired from MrFun's answer:
import glob
import pandas as pd
list_of_csv_files = glob.glob(directory_path + '/*.csv')
list_of_csv_files.sort()
df = pd.concat(map(pd.read_csv, list_of_csv_files), ignore_index=True)
Notes:
By default, the list of files generated through glob.glob is not sorted. On the other hand, in many scenarios, it's required to be sorted e.g. one may want to analyze number of sensor-frame-drops v/s timestamp.
In pd.concat command, if ignore_index=True is not specified then it reserves the original indices from each dataframes (i.e. each individual CSV file in the list) and the main dataframe looks like
timestamp id valid_frame
0
1
2
.
.
.
0
1
2
.
.
.
With ignore_index=True, it looks like:
timestamp id valid_frame
0
1
2
.
.
.
108
109
.
.
.
IMO, this is helpful when one may want to manually create a histogram of number of frame drops v/s one minutes (or any other duration) bins and want to base the calculation on very first timestamp e.g.
begin_timestamp = df['timestamp'][0]
Without, ignore_index=True, df['timestamp'][0] generates the series containing very first timestamp from all the individual dataframes, it does not give just a value.
Another one-liner with list comprehension which allows to use arguments with read_csv.
df = pd.concat([pd.read_csv(f'dir/{f}') for f in os.listdir('dir') if f.endswith('.csv')])
Alternative using the pathlib library (often preferred over os.path).
This method avoids iterative use of pandas concat()/apped().
From the pandas documentation:
It is worth noting that concat() (and therefore append()) makes a full copy of the data, and that constantly reusing this function can create a significant performance hit. If you need to use the operation over several datasets, use a list comprehension.
import pandas as pd
from pathlib import Path
dir = Path("../relevant_directory")
df = (pd.read_csv(f) for f in dir.glob("*.csv"))
df = pd.concat(df)
If multiple CSV files are zipped, you may use zipfile to read all and concatenate as below:
import zipfile
import pandas as pd
ziptrain = zipfile.ZipFile('yourpath/yourfile.zip')
train = []
train = [ pd.read_csv(ziptrain.open(f)) for f in ziptrain.namelist() ]
df = pd.concat(train)
Based on Sid's good answer.
To identify issues of missing or unaligned columns
Before concatenating, you can load CSV files into an intermediate dictionary which gives access to each data set based on the file name (in the form dict_of_df['filename.csv']). Such a dictionary can help you identify issues with heterogeneous data formats, when column names are not aligned for example.
Import modules and locate file paths:
import os
import glob
import pandas
from collections import OrderedDict
path =r'C:\DRO\DCL_rawdata_files'
filenames = glob.glob(path + "/*.csv")
Note: OrderedDict is not necessary, but it'll keep the order of files which might be useful for analysis.
Load CSV files into a dictionary. Then concatenate:
dict_of_df = OrderedDict((f, pandas.read_csv(f)) for f in filenames)
pandas.concat(dict_of_df, sort=True)
Keys are file names f and values are the data frame content of CSV files.
Instead of using f as a dictionary key, you can also use os.path.basename(f) or other os.path methods to reduce the size of the key in the dictionary to only the smaller part that is relevant.
import os
os.system("awk '(NR == 1) || (FNR > 1)' file*.csv > merged.csv")
Where NR and FNR represent the number of the line being processed.
FNR is the current line within each file.
NR == 1 includes the first line of the first file (the header), while FNR > 1 skips the first line of each subsequent file.
In case of an unnamed column issue, use this code for merging multiple CSV files along the x-axis.
import glob
import os
import pandas as pd
merged_df = pd.concat([pd.read_csv(csv_file, index_col=0, header=0) for csv_file in glob.glob(
os.path.join("data/", "*.csv"))], axis=0, ignore_index=True)
merged_df.to_csv("merged.csv")
You can do it this way also:
import pandas as pd
import os
new_df = pd.DataFrame()
for r, d, f in os.walk(csv_folder_path):
for file in f:
complete_file_path = csv_folder_path+file
read_file = pd.read_csv(complete_file_path)
new_df = new_df.append(read_file, ignore_index=True)
new_df.shape
Consider using convtools library, which provides lots of data processing primitives and generates simple ad hoc code under the hood.
It is not supposed to be faster than pandas/polars, but sometimes it can be.
e.g. you could concat csv files into one for further reuse - here's the code:
import glob
from convtools import conversion as c
from convtools.contrib.tables import Table
import pandas as pd
def test_pandas():
df = pd.concat(
(
pd.read_csv(filename, index_col=None, header=0)
for filename in glob.glob("tmp/*.csv")
),
axis=0,
ignore_index=True,
)
df.to_csv("out.csv", index=False)
# took 20.9 s
def test_convtools():
table = None
for filename in glob.glob("tmp/*.csv"):
table_ = Table.from_csv(filename, header=False)
if table is None:
table = table_
else:
table = table.chain(table_)
table.into_csv("out_convtools.csv", include_header=False)
# took 15.8 s
Of course if you just want to obtain a dataframe without writing a concatenated file, it will take 4.63 s and 10.9 s correspondingly (pandas is faster here because it doesn't need to zip columns for writing it back).
import pandas as pd
import glob
path = r'C:\DRO\DCL_rawdata_files' # use your path
file_path_list = glob.glob(path + "/*.csv")
file_iter = iter(file_path_list)
list_df_csv = []
list_df_csv.append(pd.read_csv(next(file_iter)))
for file in file_iter:
lsit_df_csv.append(pd.read_csv(file, header=0))
df = pd.concat(lsit_df_csv, ignore_index=True)
This is how you can do it using Colaboratory on Google Drive:
import pandas as pd
import glob
path = r'/content/drive/My Drive/data/actual/comments_only' # Use your path
all_files = glob.glob(path + "/*.csv")
li = []
for filename in all_files:
df = pd.read_csv(filename, index_col=None, header=0)
li.append(df)
frame = pd.concat(li, axis=0, ignore_index=True,sort=True)
frame.to_csv('/content/drive/onefile.csv')
I am working with time series data that is formatted as each row is a single instance of a ID/time/data. This means that the rows don't correspond 1 to 1 for each ID. Each ID has many rows across time.
I am trying to use dask delayed to have a function run on an entire ID sequence (it makes sense that the operation should be able to run on each individual ID at the same time since they don't affect each other). To do this I am first looping through each of the ID tags, pulling/locating all the data from that ID (with .loc in pandas, so it is a separate "mini" df), then delaying the function call on the mini df, adding a column with the delayed values and adding it to a list of all mini dfs. At the end of the for loop I want to call dask.compute() on all the mini-dfs at once but for some reason the mini df's values are still delayed. Below I will post some pseudocode about what I just tried to explain.
I have a feeling that this may not be the best way to go about this but it's what made sense at the time and I can't understand whats wrong so any help would be very much appreciated.
Here is what I am trying to do:
list_of_mini_dfs = []
for id in big_df:
curr_df = big_df.loc[big_df['id'] == id]
curr_df['new value 1'] = dask.delayed(myfunc)(args1)
curr_df['new value 2'] = dask.delayed(myfunc)(args2) #same func as previous line
list_of_mini_dfs.append(curr_df)
list_of_mini_dfs = dask.delayed(list_of_mini_dfs).compute()
Concat all mini dfs into new big df.
As you can see by the code I have to reach into my big/overall dataframe to pull out each ID's sequence of data since it is interspersed throughout the rows. I want to be able to call a delayed function on that single ID's data and then return the values from the function call into the big/overall dataframe.
Currently this method is not working, when I concat all the mini dataframes back together the two values I have delayed are still delayed, which leads me to think that it is due to the way I am delaying a function within a df and trying to compute the list of dataframes. I just can't see how to fix it.
Hopefully this was relatively clear and thank you for the help.
IIUC you are trying to do a sort of transform using dask.
import pandas as pd
import dask.dataframe as dd
import numpy as np
# generate big_df
dates = pd.date_range(start='2019-01-01',
end='2020-01-01')
l = len(dates)
out = []
for i in range(1000):
df = pd.DataFrame({"ID":[i]*l,
"date": dates,
"data0": np.random.randn(l),
"data1": np.random.randn(l)})
out.append(df)
big_df = pd.concat(out, ignore_index=True)\
.sample(frac=1)\
.reset_index(drop=True)
Now you want to apply your function fun on columns data0 and data1
Pandas
out = big_df.groupby("ID")[["data0","data1"]]\
.apply(fun)\
.reset_index()
df_pd = pd.merge(big_df, out, how="left", on="ID" )
Dask
df = dd.from_pandas(big_df, npartitions=4)
out = df.groupby("ID")[["data0","data1"]]\
.apply(fun, meta={'data0':'f8',
'data1':'f8'})\
.rename(columns={'data0': 'new_values0',
'data1': 'new_values1'})\
.compute() # Here you need to compute otherwise you'll get NaNs
df_dask = dd.merge(df, out,
how="left",
left_on=["ID"],
right_index=True)
The dask version is not necessarily faster than the pandas one. In particular if your df fits in RAM.
I am running the same functionality using Pandas API and Dask API.
I expected Dask API to be faster, but it is not.
Functionality
I cross joint 2 dataframes (pandas and dask respectively) by a 'grouping' column and then, on every single pair I compute the Levensthein distance between 2 strings.
The results is the expected, but I am concerned about the performance.
Pandas
#timeit
def pd_fuzzy_comparison(df1:DF, df2:DF, group_col:str, compare):
df = df1.merge(df2, on=bubble_col, suffixes=('_internal', '_external'))
df['score'] = df.apply(lambda d:
comp(d.company_norm_internal, d.company_norm_external), axis=1)
return df
Dask
#timeit
def dd_fuzzy_comparison(dd1:dd, dd2:dd, group_col:str, compare):
ddf = dd1.merge(dd2, on='city_norm', suffixes=('_internal', '_external'))
ddf['score'] = ddf.apply(
lambda d: ratio(d.company_norm_internal, d.company_norm_external), axis=1)
return ddf.compute()
Main
import multiprocessing
CORES = multiprocessing.cpu_count()
results = pd_fuzzy_comparison(
df1=internal_bubbles.copy(),
df2=external_bubbles.copy(),
bubble_col='city_norm',
compare=ratio )
ddata1 = dd.from_pandas(internal_bubbles.copy(), npartitions=CORES)
ddata2 = dd.from_pandas(external_bubbles.copy(), npartitions=CORES)
ddresults = dd_fuzzy_comparison(
dd1=ddata1.copy(), dd2=ddata2.copy(),
bubble_col='city-norm',
compare=ratio)
Output
'pd_fuzzy_comparison' 1122.39 ms
'dd_fuzzy_comparison' 1717.83 ms
What am I missing?
Thanks!
First, Dask isn't always faster than Pandas. If Pandas works for you, you should stick with it.
https://docs.dask.org/en/latest/best-practices.html#start-small
In your particular case you're using the df.apply method, which uses Python for loops, which will be slowish in any case. It is also GIL bound, so you will want to choose a scheduler that uses processes, like the dask.distributed or multiprocessing schedulers.
https://docs.dask.org/en/latest/scheduling.html
I have several Excel files that I need to load and pre-process before loading them into a Spark DF. I have a list of these files that need to be processed. I do something like this to read them in:
file_list_rdd = sc.emptyRDD()
for file_path in file_list:
current_file_rdd = sc.binaryFiles(file_path)
print(current_file_rdd.count())
file_list_rdd = file_list_rdd.union(current_file_rdd)
I then have some mapper function that turns file_list_rdd from a set of (path, bytes) tuples to (path, Pandas DataFrame) tuples. This allows me to use Pandas to read the Excel file and to manipulate the files so that they're uniform before making them into a Spark DataFrame.
How do I take an RDD of (file path, Pandas DF) tuples and turn it into a single Spark DF? I'm aware of functions that can do a single transformation, but not one that can do several.
My first attempt was something like this:
sqlCtx = SQLContext(sc)
def convert_pd_df_to_spark_df(item):
return sqlCtx.createDataFrame(item[0][1])
processed_excel_rdd.map(convert_pd_df_to_spark_df)
I'm guessing that didn't work because sqlCtx isn't distributed with the computation (it's a guess because the stack trace doesn't make much sense to me).
Thanks in advance for taking the time to read :).
Can be done using conversion to Arrow RecordBatches which Spark > 2.3 can process into a DF in a very efficient manner.
https://gist.github.com/linar-jether/7dd61ed6fa89098ab9c58a1ab428b2b5
This snippet monkey-patches spark to include a createFromPandasDataframesRDD method.
The createFromPandasDataframesRDD method accepts a RDD object of pandas DFs (Assumes same columns) and returns a single Spark DF.
I solved this by writing a function like this:
def pd_df_to_row(rdd_row):
key = rdd_row[0]
pd_df = rdd_row[1]
rows = list()
for index, series in pd_df.iterrows():
# Takes a row of a df, exports it as a dict, and then passes an unpacked-dict into the Row constructor
row_dict = {str(k):v for k,v in series.to_dict().items()}
rows.append(Row(**row_dict))
return rows
You can invoke it by calling something like:
processed_excel_rdd = processed_excel_rdd.flatMap(pd_df_to_row)
pd_df_to_row now has a collection of Spark Row objects. You can now say:
processed_excel_rdd.toDF()
There's probably something more efficient than the Series-> dict-> Row operation, but this got me through.
Why not make a list of the dataframes or filenames and then call union in a loop. Something like this:
If pandas dataframes:
dfs = [df1, df2, df3, df4]
sdf = None
for df in dfs:
if sdf:
sdf = sdf.union(spark.createDataFrame(df))
else:
sdf = spark.createDataFrame(df)
If filenames:
names = [name1, name2, name3, name4]
sdf = None
for name in names:
if sdf:
sdf = sdf.union(spark.createDataFrame(pd.read_excel(name))
else:
sdf = spark.createDataFrame(pd.read_excel(name))