Develop Reference

DASK: Typerrror: Column assignment doesn't support type numpy.ndarray whereas Pandas works fine

I'm using Dask to read in a 10m row csv+ and perform some calculations. So far it's proving to be 10x faster than Pandas.
I have a piece of code, below, that when used with pandas works fine, but with dask throws a type error. I am unsure of how to overcome the typerror. It seems like an array is being handed back to the dataframe/column by the select function when using dask, but not when using pandas? But I don't want to switch the whole thing back to pandas and lose the 10x performance benefit.
This answer is the result of some help of some others on Stack Overflow, however I think that question has deviated far enough from the initial question that this is altogether different. Code below.
PANDAS: Works
Time Taken excluding AndHeathSolRadFact: 40 seconds
import pandas as pd
import numpy as np
from timeit import default_timer as timer
start = timer()
df = pd.read_csv(r'C:\Users\i5-Desktop\Downloads\Weathergrids.csv')
df['DateTime'] = pd.to_datetime(df['Date'], format='%Y-%d-%m %H:%M')
df['Month'] = df['DateTime'].dt.month
df['Grass_FMC'] = (97.7+4.06*df['RH'])/(df['Temperature']+6)-0.00854*df['RH']+3000/df['Curing']-30
df["AndHeathSolRadFact"] = np.select(
[
(df['Month'].between(8,12)),
(df['Month'].between(1,2) & df['CloudCover']>30)
], #list of conditions
[1, 1], #list of results
default=0) #default if no match
print(df.head())
#print(ddf.tail())
end = timer()
print(end - start)
DASK: BROKEN
Time Taken excluding AndHeathSolRadFact: 4 seconds
import dask.dataframe as dd
import dask.multiprocessing
import dask.threaded
import pandas as pd
import numpy as np
# Dataframes implement the Pandas API
import dask.dataframe as dd
from timeit import default_timer as timer
start = timer()
ddf = dd.read_csv(r'C:\Users\i5-Desktop\Downloads\Weathergrids.csv')
ddf['DateTime'] = dd.to_datetime(ddf['Date'], format='%Y-%d-%m %H:%M')
ddf['Month'] = ddf['DateTime'].dt.month
ddf['Grass_FMC'] = (97.7+4.06*ddf['RH'])/(ddf['Temperature']+6)-0.00854*ddf['RH']+3000/ddf['Curing']-30
ddf["AndHeathSolRadFact"] = np.select(
[
(ddf['Month'].between(8,12)),
(ddf['Month'].between(1,2) & ddf['CloudCover']>30)
], #list of conditions
[1, 1], #list of results
default=0) #default if no match
print(ddf.head())
#print(ddf.tail())
end = timer()
print(end - start)
Error
---------------------------------------------------------------------------
TypeError Traceback (most recent call last)
<ipython-input-50-86c08f38bce6> in <module>
29 ], #list of conditions
30 [1, 1], #list of results
---> 31 default=0) #default if no match
32
33
~\Anaconda3\lib\site-packages\dask\dataframe\core.py in __setitem__(self, key, value)
3276 df = self.assign(**{k: value for k in key})
3277 else:
-> 3278 df = self.assign(**{key: value})
3279
3280 self.dask = df.dask
~\Anaconda3\lib\site-packages\dask\dataframe\core.py in assign(self, **kwargs)
3510 raise TypeError(
3511 "Column assignment doesn't support type "
-> 3512 "{0}".format(typename(type(v)))
3513 )
3514 if callable(v):
TypeError: Column assignment doesn't support type numpy.ndarray
Sample Weathegrids CSV
Location,Date,Temperature,RH,WindDir,WindSpeed,DroughtFactor,Curing,CloudCover
1075,2019-20-09 04:00,6.8,99.3,143.9,5.6,10.0,93.0,1.0
1075,2019-20-09 05:00,6.4,100.0,93.6,7.2,10.0,93.0,1.0
1075,2019-20-09 06:00,6.7,99.3,130.3,6.9,10.0,93.0,1.0
1075,2019-20-09 07:00,8.6,95.4,68.5,6.3,10.0,93.0,1.0
1075,2019-20-09 08:00,12.2,76.0,86.4,6.1,10.0,93.0,1.0

I just had a similar problem, and I was able to get it to work by converting the ndarray into a Dask array. I also had to ensure the number of partitions matched between the ndarray and the Dask DataFrame.

Assigning series to Dask column works.
dask_df['col'] = pd.Series(list or array)

For some reason that's not entirely clear to me yet, the above-mentioned solutions didn't work for me.
I ended up defining a function that does the column assignment per pandas dataframe and then mapping that function to all of my dask partitions.
def map_randoms(df):
df['col_rand'] = np.random.randint(0,2, size=len(df))
return df
ddf = ddf.map_partitions(map_randoms)
ddf.persist()

Can you please try with adding .any() or .all() at the end of you np.select() statement?
df["AndHeathSolRadFact"] = np.select(
[
(df['Month'].between(8,12)),
(df['Month'].between(1,2) & df['CloudCover']>30)
], #list of conditions
[1, 1], #list of results
default=0).all() #default if no match

This answer isn't elegant but is functional.
I found the select function was about 20 seconds quicker on an 11m row dataset in pandas. I also found that even if I performed the same function in dask that the result would return a numpy (pandas) array. Dask inherently cannot accept this, but it is possible to transfer dataframes between dask and pandas.
So, I get the benefit of loading and date transforms in dask (4 seconds compared to 40 seconds in pandas), the benefits of select using pandas (40 seconds compared to 60 seconds in dask), and just need to accept that I'll be using more memory.
There's little loss in time by transforming between dataframes.
Finally, I had to make sure that I cleaned up dataframes as python wasn't cleaning up memory between test runs and just kept accumulating.
import dask.dataframe as dd
import dask.multiprocessing
import dask.threaded
import pandas as pd
import numpy as np
# Dataframes implement the Pandas API
import dask.dataframe as dd
from timeit import default_timer as timer
start = timer()
ddf = dd.read_csv(r'C:\Users\i5-Desktop\Downloads\Weathergrids.csv')
#print(ddf.describe(include='all'))
#Wrangle the dates so we can interrogate them
ddf['DateTime'] = dd.to_datetime(ddf['Date'], format='%Y-%d-%m %H:%M')
ddf['Month'] = ddf['DateTime'].dt.month
#Grass Fuel Moisture Content
ddf['Grass_FMC'] = (97.7+4.06*ddf['RH'])/(ddf['Temperature']+6)-0.00854*ddf['RH']+3000/ddf['Curing']-30
#Convert to a Pandas DataFrame because dask was being slow with the select logic below
df = ddf.compute()
del [ddf]
#ddf["AndHeathSolRadFact"] = np.select(
#Solar Radiation Factor - this seems to take 32 seconds. Why?
df["AndHeathSolRadFact"] = np.select(
[
(df['Month'].between(8,12)),
(df['Month'].between(1,2) & df['CloudCover']>30)
], #list of conditions
[1, 1], #list of results
default=0) #default if no match
#Convert back to a Dask dataframe because we want that juicy parallelism
ddf2 = dd.from_pandas(df,npartitions=4)
del [df]
print(ddf2.head())
#print(ddf.tail())
end = timer()
print(end - start)
#Clean up remaining dataframes
del [[ddf2]]

EDITED:
I really have a elegant solution for you problem:-
from dask.array import from_array as fa
df.compute()['Name of you column'] = fa(the_list_you_want_to_assign_as_column)

python pandas: vectorized time series window function

I have a pandas dataframe in the following format:
'customer_id','transaction_dt','product','price','units'
1,2004-01-02,thing1,25,47
1,2004-01-17,thing2,150,8
2,2004-01-29,thing2,150,25
3,2017-07-15,thing3,55,17
3,2016-05-12,thing3,55,47
4,2012-02-23,thing2,150,22
4,2009-10-10,thing1,25,12
4,2014-04-04,thing2,150,2
5,2008-07-09,thing2,150,43
I have written the following to create two new fields indicating 30 day windows:
import numpy as np
import pandas as pd
start_date_period = pd.period_range('2004-01-01', '12-31-2017', freq='30D')
end_date_period = pd.period_range('2004-01-30', '12-31-2017', freq='30D')
def find_window_start_date(x):
window_start_date_idx = np.argmax(x < start_date_period.end_time)
return start_date_period[window_start_date_idx]
df['window_start_dt'] = df['transaction_dt'].apply(find_window_start_date)
def find_window_end_date(x):
window_end_date_idx = np.argmin(x > end_date_period.start_time)
return end_date_period[window_end_date_idx]
df['window_end_dt'] = df['transaction_dt'].apply(find_window_end_date)
Unfortunately, this is far too slow doing the row-wise apply for my application. I would greatly appreciate any tips on vectorizing these functions if possible.
EDIT:
The resultant dataframe should have this layout:
'customer_id','transaction_dt','product','price','units','window_start_dt','window_end_dt'
It does not need to be resampled or windowed in the formal sense. It just needs 'window_start_dt' and 'window_end_dt' columns to be added. The current code works, it just need to be vectorized if possible.

EDIT 2: pandas.cut is built-in:
tt=[[1,'2004-01-02',0.1,25,47],
[1,'2004-01-17',0.2,150,8],
[2,'2004-01-29',0.2,150,25],
[3,'2017-07-15',0.3,55,17],
[3,'2016-05-12',0.3,55,47],
[4,'2012-02-23',0.2,150,22],
[4,'2009-10-10',0.1,25,12],
[4,'2014-04-04',0.2,150,2],
[5,'2008-07-09',0.2,150,43]]
start_date_period = pd.date_range('2004-01-01', '12-01-2017', freq='MS')
end_date_period = pd.date_range('2004-01-30', '12-31-2017', freq='M')
df = pd.DataFrame(tt,columns=['customer_id','transaction_dt','product','price','units'])
df['transaction_dt'] = pd.Series([pd.to_datetime(sub_t[1],format='%Y-%m-%d') for sub_t in tt])
the_cut = pd.cut(df['transaction_dt'],bins=start_date_period,right=True,labels=False,include_lowest=True)
df['win_start_test'] = pd.Series([start_date_period[int(x)] if not np.isnan(x) else 0 for x in the_cut])
df['win_end_test'] = pd.Series([end_date_period[int(x)] if not np.isnan(x) else 0 for x in the_cut])
print(df.head())
win_start_test and win_end_test should be equal to their counterparts computed using your function.
The ValueError was coming from not casting x to int in the relevant line. I also added a NaN check, though it wasn't needed for this toy example.
Note the change to pd.date_range and the use of the start-of-month and end-of-month flags M and MS, as well as converting the date strings into datetime.

Tracking Error on a number of benchmarks

I'm trying to calculate tracking error for a number of different benchmarks versus a fund that I'm looking at (tracking error is defined as the standard deviation of the percent difference between the fund and benchmark). The time series for the fund and all the benchmarks are all in a data frame that I'm reading from an excel on file and what I have so far is this (with the idea that arg1 represents all the benchmarks and is then applied using applymap), but it's returning a KeyError, any suggestions?
import pandas as pd
import numpy as np
data = pd.read_excel('File_Path.xlsx')
def index_analytics(arg1):
tracking_err = np.std((data['Fund'] - data[arg1]) / data[arg1])
return tracking_err
data.applymap(index_analytics)

There are a few things that need fixed. First,applymap passes each individual value for all the columns to your calling function (index_analytics). So arg1 is the individual scalar value for all the values in your dataframe. data[arg1] is always going to return a key error unless all your values are also column names.
You also shouldn't need to use apply to do this. Assuming your benchmarks are in the same dataframe then you should be able to do something like this for each benchmark. Next time include a sample of your dataframe.
df['Benchmark1_result'] = (df['Fund'] - data['Benchmark1']) / data['Benchmark1']
And if you want to calculate all the standard deviations for all the benchmarks you can do this
# assume you have a dataframe with a list of all the benchmark columns
benchmark_columns = [list, of, benchmark, columns]
np.std((df['Fund'].values - df[benchmark_columns].values) / df['Fund'].values, axis=1)

Assuming you're following the definition of Tracking Error below:
import pandas as pd
import numpy as np
# Example DataFrame
df = pd.DataFrame({'Portfolio_Returns': [5.00, 1.67], 'Bench_Returns': [2.89, .759]})
df['Active_Return'] = df['Portfolio_Returns'] - df['Bench_Returns']
print(df.head())
list_ = df['Active_Return']
temp_ = []
for val in list_:
x = val**2
temp_.append(x)
tracking_error = np.sqrt(sum(temp_))
print(f"Tracking Error is: {tracking_error}")
Or if you want it more compact (because apparently the cool kids do it):
df = pd.DataFrame({'Portfolio_Returns': [5.00, 1.67], 'Bench_Returns': [2.89, .759]})
tracking_error = np.sqrt(sum([val**2 for val in df['Portfolio_Returns'] - df['Bench_Returns']]))
print(f"Tracking Error is: {tracking_error}")

Vectorize integration of pandas.DataFrame

I have a DataFrame of force-displacement data. The displacement array has been set to the DataFrame index, and the columns are my various force curves for different tests.
How do I calculate the work done (which is "the area under the curve")?
I looked at numpy.trapz which seems to do what I need, but I think that I can avoid looping over each column like this:
import numpy as np
import pandas as pd
forces = pd.read_csv(...)
work_done = {}
for col in forces.columns:
work_done[col] = np.trapz(forces.loc[col], forces.index))
I was hoping to create a new DataFrame of the areas under the curves rather than a dict, and thought that DataFrame.apply() or something might be appropriate but don't know where to start looking.
In short:
Can I avoid the looping?
Can I create a DataFrame of work done directly?
Thanks in advance for any help.

You could vectorize this by passing the whole DataFrame to np.trapz and specifying the axis= argument, e.g.:
import numpy as np
import pandas as pd
# some random input data
gen = np.random.RandomState(0)
x = gen.randn(100, 10)
names = [chr(97 + i) for i in range(10)]
forces = pd.DataFrame(x, columns=names)
# vectorized version
wrk = np.trapz(forces, x=forces.index, axis=0)
work_done = pd.DataFrame(wrk[None, :], columns=forces.columns)
# non-vectorized version for comparison
work_done2 = {}
for col in forces.columns:
work_done2.update({col:np.trapz(forces.loc[:, col], forces.index)})
These give the following output:
from pprint import pprint
pprint(work_done.T)
# 0
# a -24.331560
# b -10.347663
# c 4.662212
# d -12.536040
# e -10.276861
# f 3.406740
# g -3.712674
# h -9.508454
# i -1.044931
# j 15.165782
pprint(work_done2)
# {'a': -24.331559643023006,
# 'b': -10.347663159421426,
# 'c': 4.6622123535050459,
# 'd': -12.536039649161403,
# 'e': -10.276861220217308,
# 'f': 3.4067399176289994,
# 'g': -3.7126739591045541,
# 'h': -9.5084536839888187,
# 'i': -1.0449311137294459,
# 'j': 15.165781517623724}
There are a couple of other problems with your original example. col is a column name rather than a row index, so it needs to index the second dimension of your dataframe (i.e. .loc[:, col] rather than .loc[col]). Also, you have an extra trailing parenthesis on the last line.
Edit:
You could also generate the output DataFrame directly by .applying np.trapz to each column, e.g.:
work_done = forces.apply(np.trapz, axis=0, args=(forces.index,))
However, this isn't really 'proper' vectorization - you are still calling np.trapz separately on each column. You can see this by comparing the speed of the .apply version against calling np.trapz directly:
In [1]: %timeit forces.apply(np.trapz, axis=0, args=(forces.index,))
1000 loops, best of 3: 582 µs per loop
In [2]: %timeit np.trapz(forces, x=forces.index, axis=0)
The slowest run took 6.04 times longer than the fastest. This could mean that an
intermediate result is being cached
10000 loops, best of 3: 53.4 µs per loop
This isn't an entirely fair comparison, since the second version excludes the extra time taken to construct the DataFrame from the output numpy array, but this should still be smaller than the difference in time taken to perform the actual integration.

Here's how to get the cumulative integral along a dataframe column using the trapezoidal rule. Alternatively, the following creates a pandas.Series method for doing your choice of Trapezoidal, Simpson's or Romberger's rule (source):
import pandas as pd
from scipy import integrate
import numpy as np
#%% Setup Functions
def integrate_method(self, how='trapz', unit='s'):
'''Numerically integrate the time series.
#param how: the method to use (trapz by default)
#return
Available methods:
* trapz - trapezoidal
* cumtrapz - cumulative trapezoidal
* simps - Simpson's rule
* romb - Romberger's rule
See http://docs.scipy.org/doc/scipy/reference/integrate.html for the method details.
or the source code
https://github.com/scipy/scipy/blob/master/scipy/integrate/quadrature.py
'''
available_rules = set(['trapz', 'cumtrapz', 'simps', 'romb'])
if how in available_rules:
rule = integrate.__getattribute__(how)
else:
print('Unsupported integration rule: %s' % (how))
print('Expecting one of these sample-based integration rules: %s' % (str(list(available_rules))))
raise AttributeError
if how is 'cumtrapz':
result = rule(self.values)
result = np.insert(result, 0, 0, axis=0)
else:
result = rule(self.values)
return result
pd.Series.integrate = integrate_method
#%% Setup (random) data
gen = np.random.RandomState(0)
x = gen.randn(100, 10)
names = [chr(97 + i) for i in range(10)]
df = pd.DataFrame(x, columns=names)
#%% Cummulative Integral
df_cummulative_integral = df.apply(lambda x: x.integrate('cumtrapz'))
df_integral = df.apply(lambda x: x.integrate('trapz'))
df_do_they_match = df_cummulative_integral.tail(1).round(3) == df_integral.round(3)
if df_do_they_match.all().all():
print("Trapz produces the last row of cumtrapz")

What is the most efficient way to loop through dataframes with pandas?

I want to perform my own complex operations on financial data in dataframes in a sequential manner.
For example I am using the following MSFT CSV file taken from Yahoo Finance:
Date,Open,High,Low,Close,Volume,Adj Close
2011-10-19,27.37,27.47,27.01,27.13,42880000,27.13
2011-10-18,26.94,27.40,26.80,27.31,52487900,27.31
2011-10-17,27.11,27.42,26.85,26.98,39433400,26.98
2011-10-14,27.31,27.50,27.02,27.27,50947700,27.27
....
I then do the following:
#!/usr/bin/env python
from pandas import *
df = read_csv('table.csv')
for i, row in enumerate(df.values):
date = df.index[i]
open, high, low, close, adjclose = row
#now perform analysis on open/close based on date, etc..
Is that the most efficient way? Given the focus on speed in pandas, I would assume there must be some special function to iterate through the values in a manner that one also retrieves the index (possibly through a generator to be memory efficient)? df.iteritems unfortunately only iterates column by column.

The newest versions of pandas now include a built-in function for iterating over rows.
for index, row in df.iterrows():
# do some logic here
Or, if you want it faster use itertuples()
But, unutbu's suggestion to use numpy functions to avoid iterating over rows will produce the fastest code.

Pandas is based on NumPy arrays.
The key to speed with NumPy arrays is to perform your operations on the whole array at once, never row-by-row or item-by-item.
For example, if close is a 1-d array, and you want the day-over-day percent change,
pct_change = close[1:]/close[:-1]
This computes the entire array of percent changes as one statement, instead of
pct_change = []
for row in close:
pct_change.append(...)
So try to avoid the Python loop for i, row in enumerate(...) entirely, and
think about how to perform your calculations with operations on the entire array (or dataframe) as a whole, rather than row-by-row.

Like what has been mentioned before, pandas object is most efficient when process the whole array at once. However for those who really need to loop through a pandas DataFrame to perform something, like me, I found at least three ways to do it. I have done a short test to see which one of the three is the least time consuming.
t = pd.DataFrame({'a': range(0, 10000), 'b': range(10000, 20000)})
B = []
C = []
A = time.time()
for i,r in t.iterrows():
C.append((r['a'], r['b']))
B.append(time.time()-A)
C = []
A = time.time()
for ir in t.itertuples():
C.append((ir[1], ir[2]))
B.append(time.time()-A)
C = []
A = time.time()
for r in zip(t['a'], t['b']):
C.append((r[0], r[1]))
B.append(time.time()-A)
print B
Result:
[0.5639059543609619, 0.017839908599853516, 0.005645036697387695]
This is probably not the best way to measure the time consumption but it's quick for me.
Here are some pros and cons IMHO:
.iterrows(): return index and row items in separate variables, but significantly slower
.itertuples(): faster than .iterrows(), but return index together with row items, ir[0] is the index
zip: quickest, but no access to index of the row
EDIT 2020/11/10
For what it is worth, here is an updated benchmark with some other alternatives (perf with MacBookPro 2,4 GHz Intel Core i9 8 cores 32 Go 2667 MHz DDR4)
import sys
import tqdm
import time
import pandas as pd
B = []
t = pd.DataFrame({'a': range(0, 10000), 'b': range(10000, 20000)})
for _ in tqdm.tqdm(range(10)):
C = []
A = time.time()
for i,r in t.iterrows():
C.append((r['a'], r['b']))
B.append({"method": "iterrows", "time": time.time()-A})
C = []
A = time.time()
for ir in t.itertuples():
C.append((ir[1], ir[2]))
B.append({"method": "itertuples", "time": time.time()-A})
C = []
A = time.time()
for r in zip(t['a'], t['b']):
C.append((r[0], r[1]))
B.append({"method": "zip", "time": time.time()-A})
C = []
A = time.time()
for r in zip(*t.to_dict("list").values()):
C.append((r[0], r[1]))
B.append({"method": "zip + to_dict('list')", "time": time.time()-A})
C = []
A = time.time()
for r in t.to_dict("records"):
C.append((r["a"], r["b"]))
B.append({"method": "to_dict('records')", "time": time.time()-A})
A = time.time()
t.agg(tuple, axis=1).tolist()
B.append({"method": "agg", "time": time.time()-A})
A = time.time()
t.apply(tuple, axis=1).tolist()
B.append({"method": "apply", "time": time.time()-A})
print(f'Python {sys.version} on {sys.platform}')
print(f"Pandas version {pd.__version__}")
print(
pd.DataFrame(B).groupby("method").agg(["mean", "std"]).xs("time", axis=1).sort_values("mean")
)
## Output
Python 3.7.9 (default, Oct 13 2020, 10:58:24)
[Clang 12.0.0 (clang-1200.0.32.2)] on darwin
Pandas version 1.1.4
mean std
method
zip + to_dict('list') 0.002353 0.000168
zip 0.003381 0.000250
itertuples 0.007659 0.000728
to_dict('records') 0.025838 0.001458
agg 0.066391 0.007044
apply 0.067753 0.006997
iterrows 0.647215 0.019600

You can loop through the rows by transposing and then calling iteritems:
for date, row in df.T.iteritems():
# do some logic here
I am not certain about efficiency in that case. To get the best possible performance in an iterative algorithm, you might want to explore writing it in Cython, so you could do something like:
def my_algo(ndarray[object] dates, ndarray[float64_t] open,
ndarray[float64_t] low, ndarray[float64_t] high,
ndarray[float64_t] close, ndarray[float64_t] volume):
cdef:
Py_ssize_t i, n
float64_t foo
n = len(dates)
for i from 0 <= i < n:
foo = close[i] - open[i] # will be extremely fast
I would recommend writing the algorithm in pure Python first, make sure it works and see how fast it is-- if it's not fast enough, convert things to Cython like this with minimal work to get something that's about as fast as hand-coded C/C++.

You have three options:
By index (simplest):
>>> for index in df.index:
... print ("df[" + str(index) + "]['B']=" + str(df['B'][index]))
With iterrows (most used):
>>> for index, row in df.iterrows():
... print ("df[" + str(index) + "]['B']=" + str(row['B']))
With itertuples (fastest):
>>> for row in df.itertuples():
... print ("df[" + str(row.Index) + "]['B']=" + str(row.B))
Three options display something like:
df[0]['B']=125
df[1]['B']=415
df[2]['B']=23
df[3]['B']=456
df[4]['B']=189
df[5]['B']=456
df[6]['B']=12
Source: alphons.io

I checked out iterrows after noticing Nick Crawford's answer, but found that it yields (index, Series) tuples. Not sure which would work best for you, but I ended up using the itertuples method for my problem, which yields (index, row_value1...) tuples.
There's also iterkv, which iterates through (column, series) tuples.

Just as a small addition, you can also do an apply if you have a complex function that you apply to a single column:
http://pandas.pydata.org/pandas-docs/dev/generated/pandas.DataFrame.apply.html
df[b] = df[a].apply(lambda col: do stuff with col here)

As #joris pointed out, iterrows is much slower than itertuples and itertuples is approximately 100 times faster than iterrows, and I tested the speed of both methods in a DataFrame with 5 million records the result is for iterrows, it is 1200it/s, and itertuples is 120000it/s.
If you use itertuples, note that every element in the for loop is a namedtuple, so to get the value in each column, you can refer to the following example code
>>> df = pd.DataFrame({'col1': [1, 2], 'col2': [0.1, 0.2]},
index=['a', 'b'])
>>> df
col1 col2
a 1 0.1
b 2 0.2
>>> for row in df.itertuples():
... print(row.col1, row.col2)
...
1, 0.1
2, 0.2

For sure, the fastest way to iterate over a dataframe is to access the underlying numpy ndarray either via df.values (as you do) or by accessing each column separately df.column_name.values. Since you want to have access to the index too, you can use df.index.values for that.
index = df.index.values
column_of_interest1 = df.column_name1.values
...
column_of_interestk = df.column_namek.values
for i in range(df.shape[0]):
index_value = index[i]
...
column_value_k = column_of_interest_k[i]
Not pythonic? Sure. But fast.
If you want to squeeze more juice out of the loop you will want to look into cython. Cython will let you gain huge speedups (think 10x-100x). For maximum performance check memory views for cython.

Another suggestion would be to combine groupby with vectorized calculations if subsets of the rows shared characteristics which allowed you to do so.

look at last one
t = pd.DataFrame({'a': range(0, 10000), 'b': range(10000, 20000)})
B = []
C = []
A = time.time()
for i,r in t.iterrows():
C.append((r['a'], r['b']))
B.append(round(time.time()-A,5))
C = []
A = time.time()
for ir in t.itertuples():
C.append((ir[1], ir[2]))
B.append(round(time.time()-A,5))
C = []
A = time.time()
for r in zip(t['a'], t['b']):
C.append((r[0], r[1]))
B.append(round(time.time()-A,5))
C = []
A = time.time()
for r in range(len(t)):
C.append((t.loc[r, 'a'], t.loc[r, 'b']))
B.append(round(time.time()-A,5))
C = []
A = time.time()
[C.append((x,y)) for x,y in zip(t['a'], t['b'])]
B.append(round(time.time()-A,5))
B
0.46424
0.00505
0.00245
0.09879
0.00209

I believe the most simple and efficient way to loop through DataFrames is using numpy and numba. In that case, looping can be approximately as fast as vectorized operations in many cases. If numba is not an option, plain numpy is likely to be the next best option. As has been noted many times, your default should be vectorization, but this answer merely considers efficient looping, given the decision to loop, for whatever reason.
For a test case, let's use the example from #DSM's answer of calculating a percentage change. This is a very simple situation and as a practical matter you would not write a loop to calculate it, but as such it provides a reasonable baseline for timing vectorized approaches vs loops.
Let's set up the 4 approaches with a small DataFrame, and we'll time them on a larger dataset below.
import pandas as pd
import numpy as np
import numba as nb
df = pd.DataFrame( { 'close':[100,105,95,105] } )
pandas_vectorized = df.close.pct_change()[1:]
x = df.close.to_numpy()
numpy_vectorized = ( x[1:] - x[:-1] ) / x[:-1]
def test_numpy(x):
pct_chng = np.zeros(len(x))
for i in range(1,len(x)):
pct_chng[i] = ( x[i] - x[i-1] ) / x[i-1]
return pct_chng
numpy_loop = test_numpy(df.close.to_numpy())[1:]
#nb.jit(nopython=True)
def test_numba(x):
pct_chng = np.zeros(len(x))
for i in range(1,len(x)):
pct_chng[i] = ( x[i] - x[i-1] ) / x[i-1]
return pct_chng
numba_loop = test_numba(df.close.to_numpy())[1:]
And here are the timings on a DataFrame with 100,000 rows (timings performed with Jupyter's %timeit function, collapsed to a summary table for readability):
pandas/vectorized 1,130 micro-seconds
numpy/vectorized 382 micro-seconds
numpy/looped 72,800 micro-seconds
numba/looped 455 micro-seconds
Summary: for simple cases, like this one, you would go with (vectorized) pandas for simplicity and readability, and (vectorized) numpy for speed. If you really need to use a loop, do it in numpy. If numba is available, combine it with numpy for additional speed. In this case, numpy + numba is almost as fast as vectorized numpy code.
Other details:
Not shown are various options like iterrows, itertuples, etc. which are orders of magnitude slower and really should never be used.
The timings here are fairly typical: numpy is faster than pandas and vectorized is faster than loops, but adding numba to numpy will often speed numpy up dramatically.
Everything except the pandas option requires converting the DataFrame column to a numpy array. That conversion is included in the timings.
The time to define/compile the numpy/numba functions was not included in the timings, but would generally be a negligible component of the timing for any large dataframe.