This question already exists:
Group by +- margin using python
Closed 7 months ago.
I want to optimize my code where Group by +- margin using python. I want to group my Dataframe composed of 2 columns ['1', '2'] based on a margin +-1 (1) and +-10 (2)
For example, a really simplified overlook
[[273, 10],[274, 14],[275, 15]]
Expected output:
[[273, 10],[274, 14]],[[274, 14],[275, 15]]
My data is much more complex with nearly 1 million data points looking like this 652.125454455
This kind of code for example take me for ever, with no results
a = np.random.uniform(low=300, high=1800, size=(300000,))
b = np.random.uniform(low=0, high=7200, size=(300000,))
print("Random numbers were created")
df = pd.DataFrame({'1': a, '2':b})
df['id'] = df.index
1_MARGIN = 1
2_MARGIN = 10
tic = time.time()
group = []
for index, row in df.iterrows():
filtered_df = df[(row['1'] - 1_MARGIN < df['1']) & (df['1'] < row['1'] + 1_MARGIN) &
(row['2'] - 2_MARGIN < df['2']) & (df['2'] < row['2'] + 2_MARGIN)]
group.append(filtered_df[['id', '1']].values.tolist())
toc = time.time()
print(f"for loop: {str(1000*(toc-tic))} ms")
I also tried
data = df.groupby('1')['2'].apply(list).reset_index(name='irt')
but in this case there is no margin
I tried my best to understand what you wanted and I arrived at a very slow solution but at least it's a solution.
import pandas as pd
import numpy as np
a = np.random.uniform(low=300, high=1800, size=(300000,))
b = np.random.uniform(low=0, high=7200, size=(300000,))
df = pd.DataFrame({'1': a, '2':b})
dfbl1=np.sort(df['1'].apply(int).unique())
dfbl2=np.sort(df['2'].apply(int).unique())
MARGIN1 = 1
MARGIN2 = 10
marg1array=np.array(range(dfbl1[0],dfbl1[-1],MARGIN1))
marg2array=np.array(range(dfbl2[0],dfbl2[-1],MARGIN2))
a=time.perf_counter()
groupmarg1=[]
groupmarg2=[]
for low,upper in zip(marg1array[:-1],marg1array[1:]):
for low2,upper2 in zip(marg2array[:-1],marg2array[1:]):
groupmarg1.append(df.loc[(df['1']>low) & (df['1']<upper)&(df['2']>low2) & (df['2']<upper2)].values.tolist())
print(time.perf_counter()-a)
I also tried to do each loop seperately and intersect them which should be faster but since we're storing .values.tolist() I couldn't figure out a faster way than below.
a=time.perf_counter()
groupmarg1=[]
groupmarg2=[]
for low,upper in zip(marg1array[:-1],marg1array[1:]):
groupmarg1.append(df.loc[(df['1']>low) & (df['1']<upper)])
newgroup=[]
for subgroup in groupmarg1:
for low2,upper2 in zip(marg2array[:-1],marg2array[1:]):
newgroup.append(subgroup.loc[(subgroup['2']>low2) & (subgroup['2']<upper2)].values.tolist())
print(time.perf_counter()-a)
which runs in ~9mins on my machine.
Oh and you need to filter out the empty dataframes, and if you want them as values.tolist() you can do it while filtering like this
gr2=[grp.values.tolist() for grp in newgroup if not grp.empty]
Related
I was trying to find the most efficient way to do range selection in a pandas dataframe on a String index (In my real world use case, those are dates that I have converted to string : would it be faster with pandas Datetime ?)
Setup
import pandas as pd
import io
import time as t
df = pd.read_csv(io.StringIO(
"col,value\n\
a,1\n\
b,2\n\
c,3\n\
d,1\n\
e,2\n\
f,3\n\
g,1\n\
h,2\n\
i,3\n\
j,1\n\
k,2\n\
l,3\n\
m,1\n\
n,2\n\
p,3"
))
df.index = df['col']
num_runs=1000
And the 3 tests :
# solution 1
start = t.process_time()
for i in range(0,num_runs):
df.loc["a":"j"]
end = t.process_time()
print(f"1: {1e4*(end-start)/num_runs:.2f}m")
#solution 2
start = t.process_time()
for i in range(0,num_runs):
df[(df.index >= "a")& (df.index <= "j" )]
end = t.process_time()
print(f"2: {1e4*(end-start)/num_runs:.2f}m")
#solution 3
start = t.process_time()
for i in range(0,num_runs):
df[(df["col"] >= "a") & (df["col"] <= "j")]
end = t.process_time()
print(f"3: {1e4*(end-start)/num_runs:.2f}m")
The output is :
1: 1.56m
2: 5.00m
3: 8.75m
I understand why 3 is much slower than the other 2 but I'm not sure I understand why 1 and 2 are that much different.
Any python/pandas guru to shed some light ?
Also if there is an even more efficient way I'm all ears of course.
I am trying to improve the performance of a current piece of code, whereby I loop through a dataframe (dataframe 'r') and find the average values from another dataframe (dataframe 'p') based on criteria.
I want to find the average of all values (column 'Val') from dataframe 'p' where (r.RefDate = p.RefDate) & (r.Item = p.Item) & (p.StartDate >= r.StartDate) & (p.EndDate <= r.EndDate)
Dummy data for this can be generated as per the below;
import pandas as pd
import numpy as np
from datetime import datetime
######### START CREATION OF DUMMY DATA ##########
rng = pd.date_range('2019-01-01', '2019-10-28')
daily_range = pd.date_range('2019-01-01','2019-12-31')
p = pd.DataFrame(columns=['RefDate','Item','StartDate','EndDate','Val'])
for item in ['A','B','C','D']:
for date in daily_range:
daily_p = pd.DataFrame({ 'RefDate': rng,
'Item':item,
'StartDate':date,
'EndDate':date,
'Val' : np.random.randint(0,100,len(rng))})
p = p.append(daily_p)
r = pd.DataFrame(columns=['RefDate','Item','PeriodStartDate','PeriodEndDate','AvgVal'])
for item in ['A','B','C','D']:
r1 = pd.DataFrame({ 'RefDate': rng,
'Item':item,
'PeriodStartDate':'2019-10-25',
'PeriodEndDate':'2019-10-31',#datetime(2019,10,31),
'AvgVal' : 0})
r = r.append(r1)
r.reset_index(drop=True,inplace=True)
######### END CREATION OF DUMMY DATA ##########
The piece of code I currently have calculating and would like to improve the performance of is as follows
for i in r.index:
avg_price = p['Val'].loc[((p['StartDate'] >= r.loc[i]['PeriodStartDate']) &
(p['EndDate'] <= r.loc[i]['PeriodEndDate']) &
(p['RefDate'] == r.loc[i]['RefDate']) &
(p['Item'] == r.loc[i]['Item']))].mean()
r['AvgVal'].loc[i] = avg_price
The first change is that generating r DataFrame, both PeriodStartDate and
PeriodEndDate are created as datetime, see the following fragment of your
initiation code, changed by me:
r1 = pd.DataFrame({'RefDate': rng, 'Item':item,
'PeriodStartDate': pd.to_datetime('2019-10-25'),
'PeriodEndDate': pd.to_datetime('2019-10-31'), 'AvgVal': 0})
To get better speed, I the set index in both DataFrames to RefDate and Item
(both columns compared on equality) and sorted by index:
p.set_index(['RefDate', 'Item'], inplace=True)
p.sort_index(inplace=True)
r.set_index(['RefDate', 'Item'], inplace=True)
r.sort_index(inplace=True)
This way, the access by index is significantly quicker.
Then I defined the following function computing the mean for rows
from p "related to" the current row from r:
def myMean(row):
pp = p.loc[row.name]
return pp[pp.StartDate.ge(row.PeriodStartDate) &
pp.EndDate.le(row.PeriodEndDate)].Val.mean()
And the only thing to do is to apply this function (to each row in r) and
save the result in AvgVal:
r.AvgVal = r.apply(myMean2, axis=1)
Using %timeit, I compared the execution time of the code proposed by EdH with mine
and got the result almost 10 times shorter.
Check on your own.
By using iterrows I managed to improve the performance, although still may be quicker ways.
for index, row in r.iterrows():
avg_price = p['Val'].loc[((p['StartDate'] >= row.PeriodStartDate) &
(p['EndDate'] <= row.PeriodEndDate) &
(p['RefDate'] == row.RefDate) &
(p['Item'] == row.Item))].mean()
r.loc[index, 'AvgVal'] = avg_price
I'm working with a relatively large dataset (approx 5m observations, made up of about 5.5k firms).
I needed to run OLS regressions with a 60 month rolling window for each firm. I noticed that the performance was insanely slow when I ran the following code:
for idx, sub_df in master_df.groupby("firm_id"):
# OLS code
However, when I first split my dataframe into about 5.5k dfs and then iterated over each of the dfs, the performance improved dramatically.
grouped_df = master_df.groupby("firm_id")
df_list = [group for group in grouped_df]
for df in df_list:
my_df = df[1]
# OLS code
I'm talking 1-2 weeks of time (24/7) to complete in the first version compared to 8-9 hours tops.
Can anyone please explain why splitting the master df into N smaller dfs and then iterating over each smaller df performs better than iterating over the same number of groups within the master df?
Thanks ever so much!
I'm unable to reproduce your observation. Here's some code that generates data and then times the direct and indirect methods separately. The time taken is very similar in either case.
Is it possible that you accidentally sorted the dataframe by the group key between the runs? Sorting by group key results in a noticeable difference in run time.
Otherwise, I'm beginning to think that there might be some other differences in your code. It would be great if you could post the full code.
import numpy as np
import pandas as pd
from datetime import datetime
def generate_data():
''' returns a Pandas DF with columns 'firm_id' and 'score' '''
# configuration
np.random.seed(22)
num_groups = 50000 # number of distinct groups in the DF
mean_group_length = 200 # how many records per group?
cov_group_length = 0.10 # throw in some variability in the num records per group
# simulate group lengths
stdv_group_length = mean_group_length * cov_group_length
group_lengths = np.random.normal(
loc=mean_group_length,
scale=stdv_group_length,
size=(num_groups,)).astype(int)
group_lengths[group_lengths <= 0] = mean_group_length
# final length of DF
total_length = sum(group_lengths)
# compute entries for group key column
firm_id_list = []
for i, l in enumerate(group_lengths):
firm_id_list.extend([(i + 1)] * l)
# construct the DF; data column is 'score' populated with Numpy's U[0, 1)
result_df = pd.DataFrame(data={
'firm_id': firm_id_list,
'score': np.random.rand(total_length)
})
# Optionally, shuffle or sort the DF by group keys
# ALTERNATIVE 1: (badly) unsorted df
result_df = result_df.sample(frac=1, random_state=13).reset_index(drop=True)
# ALTERNATIVE 2: sort by group key
# result_df.sort_values(by='firm_id', inplace=True)
return result_df
def time_method(df, method):
''' time 'method' with 'df' as its argument '''
t_start = datetime.now()
method(df)
t_final = datetime.now()
delta_t = t_final - t_start
print(f"Method '{method.__name__}' took {delta_t}.")
return
def process_direct(df):
''' direct for-loop over groupby object '''
for group, df in df.groupby('firm_id'):
m = df.score.mean()
s = df.score.std()
return
def process_indirect(df):
''' indirect method: generate groups first as list and then loop over list '''
grouped_df = df.groupby('firm_id')
group_list = [pair for pair in grouped_df]
for pair in group_list:
m = pair[1].score.mean()
s = pair[1].score.std()
df = generate_data()
time_method(df, process_direct)
time_method(df, process_indirect)
I know that a few posts have been made regarding how to output the unique values of a dataframe without reordering the data.
I have tried many times to implement these methods, however, I believe that the problem relates to how the dataframe in question has been defined.
Basically, I want to look into the dataframe named "C", and output the unique values into a new dataframe named "C1", without changing the order in which they are stored at the moment.
The line that I use currently is:
C1 = pd.DataFrame(np.unique(C))
However, this returns an ascending order list (while, I simply want the list order preserved only with duplicates removed).
Once again, I apologise to the advanced users who will look at my code and shake their heads -- I'm still learning! And, yes, I have tried numerous methods to solve this problem (redefining the C dataframe, converting the output to be a list etc), to no avail unfortunately, so this is my cry for help to the Python gods. I defined both C and C1 as dataframes, as I understand that these are pretty much the best datastructures to house data in, such that they can be recalled and used later, plus it is quite useful to name the columns without affecting the data contained in the dataframe).
Once again, your help would be much appreciated.
F0 = ('08/02/2018','08/02/2018',50)
F1 = ('08/02/2018','09/02/2018',52)
F2 = ('10/02/2018','11/02/2018',46)
F3 = ('12/02/2018','16/02/2018',55)
F4 = ('09/02/2018','28/02/2018',48)
F_mat = [[F0,F1,F2,F3,F4]]
F_test = pd.DataFrame(np.array(F_mat).reshape(5,3),columns=('startdate','enddate','price'))
#convert string dates into DateTime data type
F_test['startdate'] = pd.to_datetime(F_test['startdate'])
F_test['enddate'] = pd.to_datetime(F_test['enddate'])
#convert datetype to be datetime type for columns startdate and enddate
F['startdate'] = pd.to_datetime(F['startdate'])
F['enddate'] = pd.to_datetime(F['enddate'])
#create contract duration column
F['duration'] = (F['enddate'] - F['startdate']).dt.days + 1
#re-order the F matrix by column 'duration', ensure that the bootstrapping
#prioritises the shorter term contracts
F.sort_values(by=['duration'], ascending=[True])
# create prices P
P = pd.DataFrame()
for index, row in F.iterrows():
new_P_row = pd.Series()
for date in pd.date_range(row['startdate'], row['enddate']):
new_P_row[date] = row['price']
P = P.append(new_P_row, ignore_index=True)
P.fillna(0, inplace=True)
#create C matrix, which records the unique day prices across the observation interval
C = pd.DataFrame(np.zeros((1, intNbCalendarDays)))
C.columns = tempDateRange
#create the Repatriation matrix, which records the order in which contracts will be
#stored in the A matrix, which means that once results are generated
#from the linear solver, we know exactly which CalendarDays map to
#which columns in the results array
#this array contains numbers from 1 to NbContracts
R = pd.DataFrame(np.zeros((1, intNbCalendarDays)))
R.columns = tempDateRange
#define a zero filled matrix, P1, which will house the dominant daily prices
P1 = pd.DataFrame(np.zeros((intNbContracts, intNbCalendarDays)))
#rename columns of P1 to be the dates contained in matrix array D
P1.columns = tempDateRange
#create prices in correct rows in P
for i in list(range(0, intNbContracts)):
for j in list(range(0, intNbCalendarDays)):
if (P.iloc[i, j] != 0 and C.iloc[0,j] == 0) :
flUniqueCalendarMarker = P.iloc[i, j]
C.iloc[0,j] = flUniqueCalendarMarker
P1.iloc[i,j] = flUniqueCalendarMarker
R.iloc[0,j] = i
for k in list(range(j+1,intNbCalendarDays)):
if (C.iloc[0,k] == 0 and P.iloc[i,k] != 0):
C.iloc[0,k] = flUniqueCalendarMarker
P1.iloc[i,k] = flUniqueCalendarMarker
R.iloc[0,k] = i
elif (C.iloc[0,j] != 0 and P.iloc[i,j] != 0):
P1.iloc[i,j] = C.iloc[0,j]
#convert C dataframe into C_list, in prepataion for converting C_list
#into a unique, order preserved list
C_list = C.values.tolist()
#create C1 matrix, which records the unique day prices across unique days in the observation period
C1 = pd.DataFrame(np.unique(C))
Use DataFrame.duplicated() to check if your data-frame contains any duplicate or not.
If yes then you can try DataFrame.drop_duplicate() .
I have a large dataset stored as a pandas panel. I would like to count the occurence of values < 1.0 on the minor_axis for each item in the panel. What I have so far:
#%% Creating the first Dataframe
dates1 = pd.date_range('2014-10-19','2014-10-20',freq='H')
df1 = pd.DataFrame(index = dates)
n1 = len(dates)
df1.loc[:,'a'] = np.random.uniform(3,10,n1)
df1.loc[:,'b'] = np.random.uniform(0.9,1.2,n1)
#%% Creating the second DataFrame
dates2 = pd.date_range('2014-10-18','2014-10-20',freq='H')
df2 = pd.DataFrame(index = dates2)
n2 = len(dates2)
df2.loc[:,'a'] = np.random.uniform(3,10,n2)
df2.loc[:,'b'] = np.random.uniform(0.9,1.2,n2)
#%% Creating the panel from both DataFrames
dictionary = {}
dictionary['First_dataset'] = df1
dictionary['Second dataset'] = df2
P = pd.Panel.from_dict(dictionary)
#%% I want to count the number of values < 1.0 for all datasets in the panel
## Only for minor axis b, not minor axis a, stored seperately for each dataset
for dataset in P:
P.loc[dataset,:,'b'] #I need to count the numver of values <1.0 in this pandas_series
To count all the "b" values < 1.0, I would first isolate b in its own DataFrame by swapping the minor axis and the items.
In [43]: b = P.swapaxes("minor","items").b
In [44]: b.where(b<1.0).stack().count()
Out[44]: 30
Thanks for thinking with me guys, but I managed to figure out a surprisingly easy solution after many hours of attempting. I thought I should share it in case someone else is looking for a similar solution.
for dataset in P:
abc = P.loc[dataset,:,'b']
abc_low = sum(i < 1.0 for i in abc)