Python optimization of loop in data frame with max and min values - pandas

I have question how can I optimize my code, in fact only the loops. I use to calculate solutions maximum of two rows, or sometimes max of row and number.
I tried to change my code using .loc and .clip but when it is about max or min which shows up multiple times I have some troubles with logical expressions.
That it was looking at the begining:
def Calc(row):
if row['Forecast'] == 0:
return max(row['Qty'],0)
elif row['def'] == 1:
return 0
elif row['def'] == 0:
return round(max(row['Qty'] - ( max(row['Forecast_total']*14,(row['Qty_12m_1']+row['Qty_12m_2'])) * max(1, (row['Total']/row['Forecast'])/54)),0 ))
df['Calc'] = df.apply(Calc, axis=1)
I menaged to change it using functions that I pointed but I have a problem how to write this max(max())
df.loc[(combined_sf2['Forecast'] == 0),'Calc'] = df.clip(0,None)
df.loc[(combined_sf2['def'] == 1),'Calc'] = 0
df.loc[(combined_sf2['def'] == 0),'Calc'] = round(max(df['Qty']- (max(df['Forecast_total']
*14,(df['Qty_12m_1']+df['Qty_12m_2']))
*max(1, (df['Total']/df['Forecast'])/54)),0))
First two functions are working, the last one doesn't.
id Forecast def Calc Qty Forecast_total Qty_12m_1 Qty_12m_2 Total
31551 0 0 0 2 0 0 0 95
27412 0,1 0 1 3 0,1 11 0 7
23995 0,1 0 0 4 0 1 0 7
27411 5,527 1 0,036186 60 0,2 64 0 183
28902 5,527 0 0,963814 33 5,327 277 0 183
23954 5,527 0 0 6 0 6 0 183
23994 5,527 0 0 8 0 0 0 183
31549 5,527 0 0 6 0 1 0 183
31550 5,527 0 0 6 0 10 0 183


Use numpy.select and instead max use numpy.maximum:
m1 = df['Forecast'] == 0
m2 = df['def'] == 1
m3 = df['def'] == 0
s1 = df['Qty'].clip(lower=0)
s3 = round(np.maximum(df['Qty'] - (np.maximum(df['Forecast_total']*14,(df['Qty_12m_1']+df['Qty_12m_2'])) * np.maximum(1, (df['Total']/df['Forecast'])/54)),0 ))
df['Calc2'] = np.select([m1, m2, m3], [s1, 0, s3], default=None)

Related

Display commend in ampl

I have a 2 dimension variable in ampl and I want to display it. I want to change the order of the indices but I do not know how to do that! I put my code , data and out put I described what kind of out put I want to have.
Here is my code:
param n;
param t;
param w;
param p;
set Var, default{1..n};
set Ind, default{1..t};
set mode, default{1..w};
var E{mode, Ind};
var B{mode,Var};
var C{mode,Ind};
param X{mode,Var,Ind};
var H{Ind};
minimize obj: sum{m in mode,i in Ind}E[m,i];
s.t. a1{m in mode, i in Ind}: sum{j in Var} X[m,j,i]*B[m,j] -C[m,i] <=E[m,i];
solve;
display C;
data;
param w:=4;
param n:=9;
param t:=2;
param X:=
[*,*,1]: 1 2 3 4 5 6 7 8 9 :=
1 69 59 100 70 35 1 1 0 0
2 34 31 372 71 35 1 0 1 0
3 35 25 417 70 35 1 0 0 1
4 0 10 180 30 35 1 0 0 0
[*,*,2]: 1 2 3 4 5 6 7 8 9 :=
1 64 58 68 68 30 2 1 0 0
2 44 31 354 84 30 2 0 1 0
3 53 25 399 85 30 2 0 0 1
4 0 11 255 50 30 2 0 0 0
The output of this code using glpksol is like tis:
C[1,1].val = -1.11111111111111
C[1,2].val = -1.11111111111111
C[2,1].val = -0.858585858585859
C[2,2].val = -1.11111111111111
C[3,1].val = -0.915032679738562
C[3,2].val = -1.11111111111111
C[4,1].val = 0.141414141414141
C[4,2].val = 0.2003367003367
but I want the result to be like this:
C[1,1].val = -1.11111111111111
C[2,1].val = -0.858585858585859
C[3,1].val = -0.915032679738562
C[4,1].val = 0.141414141414141
C[1,2].val = -1.11111111111111
C[2,2].val = -1.11111111111111
C[3,2].val = -1.11111111111111
C[4,2].val = 0.2003367003367
any idea?

You can use for loops and printf commands in your .run file:
for {i in Ind}
for {m in mode}
printf "C[%d,%d] = %.4f\n", m, i, C[m,i];
or even:
printf {i in Ind, m in mode} "C[%d,%d] = %.4f\n", m, i, C[m,i];
I don't get the same numerical results as you, but anyway the output works:
C[1,1] = 0.0000
C[2,1] = 0.0000
C[3,1] = 0.0000
C[4,1] = 0.0000
C[1,2] = 0.0000
C[2,2] = 0.0000
C[3,2] = 0.0000
C[4,2] = 0.0000

How to speed up Pandas' "iterrows"

I have a Pandas dataframe which I want to transform in the following way: I have some sensor data from an intelligent floor which is in column "CAPACITANCE" (split by ",") and that data comes from the device indicated in column "DEVICE". Now I want to have one row with a column per sensor - each device has 8 sensors, so I want to have devices x 8 columns and in that column I want the sensor data from exactly that sensor.
But my code seems to be super slow since I have about 90.000 rows in that dataframe! Does anyone have a suggestion how to speed it up?
BEFORE:
CAPACITANCE DEVICE TIMESTAMP \
0 0.00,-1.00,0.00,1.00,1.00,-2.00,13.00,1.00 01,07 2017/11/15 12:24:42
1 0.00,0.00,-1.00,-1.00,-1.00,0.00,-1.00,0.00 01,07 2017/11/15 12:24:42
2 0.00,-1.00,-2.00,0.00,0.00,1.00,0.00,-2.00 01,07 2017/11/15 12:24:43
3 2.00,0.00,-2.00,-1.00,0.00,0.00,1.00,-2.00 01,07 2017/11/15 12:24:43
4 1.00,0.00,-2.00,1.00,1.00,-3.00,5.00,1.00 01,07 2017/11/15 12:24:44
AFTER:
01,01-0 01,01-1 01,01-2 01,01-3 01,01-4 01,01-5 01,01-6 01,01-7 \
0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0
2 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0
01,02-0 01,02-1 ... 05,07-1 05,07-2 05,07-3 05,07-4 05,07-5 \
0 0 0 ... 0 0 0 0 0
1 0 0 ... 0 0 0 0 0
2 0 0 ... 0 0 0 0 0
3 0 0 ... 0 0 0 0 0
4 0 0 ... 0 0 0 0 0
05,07-6 05,07-7 TIMESTAMP 01,07-8
0 0 0 2017-11-15 12:24:42 1.00
1 0 0 2017-11-15 12:24:42 0.00
2 0 0 2017-11-15 12:24:43 -2.00
3 0 0 2017-11-15 12:24:43 -2.00
4 0 0 2017-11-15 12:24:44 1.00
# creating new dataframe based on the old one
floor_df_resampled = floor_df.copy()
floor_device = ["01,01", "01,02", "01,03", "01,04", "01,05", "01,06", "01,07", "01,08", "01,09", "01,10",
"02,01", "02,02", "02,03", "02,04", "02,05", "02,06", "02,07", "02,08", "02,09", "02,10",
"03,01", "03,02", "03,03", "03,04", "03,05", "03,06", "03,07", "03,08", "03,09",
"04,01", "04,02", "04,03", "04,04", "04,05", "04,06", "04,07", "04,08", "04,09",
"05,06", "05,07"]
# creating new columns
floor_objects = []
for device in floor_device:
for sensor in range(8):
floor_objects.append(device + "-" + str(sensor))
# merging new columns
floor_df_resampled = pd.concat([floor_df_resampled, pd.DataFrame(columns=floor_objects)], ignore_index=True, sort=True)
# part that takes loads of time
for index, row in floor_df_resampled.iterrows():
obj = row["DEVICE"]
sensor_data = row["CAPACITANCE"].split(',')
for idx, val in enumerate(sensor_data):
col = obj + "-" + str(idx + 1)
floor_df_resampled.loc[index, col] = val
floor_df_resampled.drop(["DEVICE"], axis=1, inplace=True)
floor_df_resampled.drop(["CAPACITANCE"], axis=1, inplace=True)

Like commented, I'm not sure why you want that many columns, but the new columns can be created as follows:
def explode(x):
dev_name = x.DEVICE.iloc[0]
ret_df = x.CAPACITANCE.str.split(',', expand=True).astype(float)
ret_df.columns = [f'{dev_name}-{col}' for col in ret_df.columns]
return ret_df
new_df = df.groupby('DEVICE').apply(explode).fillna(0)
and then you can merge this with the old data frame:
df = df.join(new_df)

how to convert pandas dataframe to libsvm format?

I have pandas data frame like below.
df
Out[50]:
0 1 2 3 4 5 6 7 8 9 ... 90 91 92 93 94 95 96 97 \
0 0 0 0 0 0 0 0 0 0 0 ... 1 1 1 1 1 1 1 1
1 0 1 1 1 0 0 1 1 1 1 ... 0 0 0 0 0 0 0 0
2 1 1 1 1 1 1 1 1 1 1 ... 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0 0 0 ... 1 1 1 1 1 1 1 1
4 0 0 0 0 0 0 0 0 0 0 ... 1 1 1 1 1 1 1 1
5 1 0 0 1 1 1 1 0 0 0 ... 0 0 0 0 0 0 0 0
6 0 0 0 0 0 0 0 0 0 0 ... 1 1 1 1 1 1 1 1
7 0 0 0 0 0 0 0 0 0 0 ... 1 1 1 1 1 1 1 1
[8 rows x 100 columns]
I have target variable as an array as below.
[1, -1, -1, 1, 1, -1, 1, 1]
How can I map this target variable to a data frame and convert it into lib SVM format?.
equi = {0:1, 1:-1, 2:-1,3:1,4:1,5:-1,6:1,7:1}
df["labels"] = df.index.map[(equi)]
d = df[np.setdiff1d(df.columns,['indx','labels'])]
e = df.label
dump_svmlight_file(d,e,'D:/result/smvlight2.dat')er code here
ERROR:
File "D:/spyder/april.py", line 54, in <module>
df["labels"] = df.index.map[(equi)]
TypeError: 'method' object is not subscriptable
When I use
df["labels"] = df.index.list(map[(equi)])
ERROR:
AttributeError: 'RangeIndex' object has no attribute 'list'
Please help me to solve those errors.

I think you need convert index to_series and then call map:
df["labels"] = df.index.to_series().map(equi)
Or use rename of index:
df["labels"] = df.rename(index=equi).index
All together:
For difference of columns pandas has difference:
from sklearn.datasets import dump_svmlight_file
equi = {0:1, 1:-1, 2:-1,3:1,4:1,5:-1,6:1,7:1}
df["labels"] = df.rename(index=equi).index
e = df["labels"]
d = df[df.columns.difference(['indx','labels'])]
dump_svmlight_file(d,e,'C:/result/smvlight2.dat')
Also it seems label column is not necessary:
from sklearn.datasets import dump_svmlight_file
equi = {0:1, 1:-1, 2:-1,3:1,4:1,5:-1,6:1,7:1}
e = df.rename(index=equi).index
d = df[df.columns.difference(['indx'])]
dump_svmlight_file(d,e,'C:/result/smvlight2.dat')

How to set (1) to max elements in pandas dataframe and (0) to everything else?

Let's say I have a pandas DataFrame.
df = pd.DataFrame(index = [ix for ix in range(10)], columns=list('abcdef'), data=np.random.randn(10,6))
df:
a b c d e f
0 -1.238393 -0.755117 -0.228638 -0.077966 0.412947 0.887955
1 -0.342087 0.296171 0.177956 0.701668 -0.481744 -1.564719
2 0.610141 0.963873 -0.943182 -0.341902 0.326416 0.818899
3 -0.561572 0.063588 -0.195256 -1.637753 0.622627 0.845801
4 -2.506322 -1.631023 0.506860 0.368958 1.833260 0.623055
5 -1.313919 -1.758250 -1.082072 1.266158 0.427079 -1.018416
6 -0.781842 1.270133 -0.510879 -1.438487 -1.101213 -0.922821
7 -0.456999 0.234084 1.602635 0.611378 -1.147994 1.204318
8 0.497074 0.412695 -0.458227 0.431758 0.514382 -0.479150
9 -1.289392 -0.218624 0.122060 2.000832 -1.694544 0.773330
how to I get set 1 to rowwise max and 0 to other elements?
I came up with:
>>> for i in range(len(df)):
... df.loc[i][df.loc[i].idxmax(axis=1)] = 1
... df.loc[i][df.loc[i] != 1] = 0
generates
df:
a b c d e f
0 0 0 0 0 0 1
1 0 0 0 1 0 0
2 0 1 0 0 0 0
3 0 0 0 0 0 1
4 0 0 0 0 1 0
5 0 0 0 1 0 0
6 0 1 0 0 0 0
7 0 0 1 0 0 0
8 0 0 0 0 1 0
9 0 0 0 1 0 0
Does anyone has a better way of doing it? May be by getting rid of the for loop or applying lambda?

Use max and check for equality using eq and cast the boolean df to int using astype, this will convert True and False to 1 and 0:
In [21]:
df = pd.DataFrame(index = [ix for ix in range(10)], columns=list('abcdef'), data=np.random.randn(10,6))
df
Out[21]:
a b c d e f
0 0.797000 0.762125 -0.330518 1.117972 0.817524 0.041670
1 0.517940 0.357369 -1.493552 -0.947396 3.082828 0.578126
2 1.784856 0.672902 -1.359771 -0.090880 -0.093100 1.099017
3 -0.493976 -0.390801 -0.521017 1.221517 -1.303020 1.196718
4 0.687499 -2.371322 -2.474101 -0.397071 0.132205 0.034631
5 0.573694 -0.206627 -0.106312 -0.661391 -0.257711 -0.875501
6 -0.415331 1.185901 1.173457 0.317577 -0.408544 -1.055770
7 -1.564962 -0.408390 -1.372104 -1.117561 -1.262086 -1.664516
8 -0.987306 0.738833 -1.207124 0.738084 1.118205 -0.899086
9 0.282800 -1.226499 1.658416 -0.381222 1.067296 -1.249829
In [22]:
df = df.eq(df.max(axis=1), axis=0).astype(int)
df
Out[22]:
a b c d e f
0 0 0 0 1 0 0
1 0 0 0 0 1 0
2 1 0 0 0 0 0
3 0 0 0 1 0 0
4 1 0 0 0 0 0
5 1 0 0 0 0 0
6 0 1 0 0 0 0
7 0 1 0 0 0 0
8 0 0 0 0 1 0
9 0 0 1 0 0 0
Timings
In [24]:
# #Raihan Masud's method
%timeit df.apply( lambda x: np.where(x == x.max() , 1 , 0) , axis = 1)
# mine
%timeit df.eq(df.max(axis=1), axis=0).astype(int)
100 loops, best of 3: 7.94 ms per loop
1000 loops, best of 3: 640 µs per loop
In [25]:
# #Nader Hisham's method
%%timeit
def max_binary(df):
binary = np.where( df == df.max() , 1 , 0 )
return binary
​
df.apply( max_binary , axis = 1)
100 loops, best of 3: 9.63 ms per loop
You can see that my method is over 12X faster than #Raihan's method
In [4]:
%%timeit
for i in range(len(df)):
df.loc[i][df.loc[i].idxmax(axis=1)] = 1
df.loc[i][df.loc[i] != 1] = 0
10 loops, best of 3: 21.1 ms per loop
The for loop is also significantly slower

import numpy as np
def max_binary(df):
binary = np.where( df == df.max() , 1 , 0 )
return binary
df.apply( max_binary , axis = 1)

Following Nader's pattern, this is a shorter version:
df.apply( lambda x: np.where(x == x.max() , 1 , 0) , axis = 1)

Complex Excel Formula in Pandas

Excel Formulas I am trying to replicate in pandas:
Click here to download workbook
* Look at columns D, E and F
entsig and exsig are manual and can be changed. In real life they would be derived from the value of another column or a comparison of two other columns
ent = 1 if entsig previous = 1 and in = 0
in = 1 if ent previous = 1 or (in previous = 1 and ex = 0)
ex = 1 if exsig previous = 1 and in previous = 1
so either ent, in, or ex will always be = 1 but never more than one of them
import pandas as pd
df = pd.DataFrame(
[[0,0,0,0,0], [1,0,0,0,0], [1,0,0,0,0], [1,0,0,0,0], [0,0,0,0,0],
[0,1,0,0,0], [0,1,0,0,0], [1,0,0,0,0], [1,0,0,0,0], [0,0,0,0,0],
[0,0,0,0,0], [0,0,0,0,0], [0,1,0,0,0], [0,1,0,0,0], [0,1,0,0,0],
[0,0,0,0,0], [0,0,0,0,0], [1,0,0,0,0], [1,0,0,0,0], [1,0,0,0,0],
[1,1,0,0,0], [0,1,0,0,0], [0,1,0,0,0], [0,1,0,0,0]],
columns=['entsig', 'exsig','ent', 'in', 'ex'])
for i in df.index:
df['ent'][(df.entsig.shift(1)==1) & (df['ent'].shift(1) == 0) & (df['in'].shift(1) == 0)]=1
df['ex'][(df.exsig.shift(1)==1) & (df['in'].shift(1)==1)]=1
df['in'][(df.ent.shift(1)==1) | ((df['in'].shift(1)==1) & (df['ex']==0))]=1
for j in df.index:
df['ent'][df['in'] == 1]=0
df['in'][df['ex']==1]=0
df['ex'][df['ex'].shift(1)==1]=0
df
results in
entsig exsig ent in ex
0 0 0 0 0 0
1 1 0 0 0 0
2 1 0 1 0 0
3 1 0 0 1 0
4 0 0 0 1 0
5 0 1 0 1 0
6 0 1 0 0 1
7 1 0 0 0 0
8 1 0 1 0 0
9 0 0 0 1 0
10 0 0 0 1 0
11 0 0 0 1 0
12 0 1 0 1 0
13 0 1 0 0 1
14 0 1 0 0 0
15 0 0 0 0 0
16 0 0 0 0 0
17 1 0 0 0 0
18 1 0 1 0 0
19 1 0 0 1 0
20 1 1 0 1 0
21 0 1 0 0 1
22 0 1 0 0 0
23 0 1 0 0 0
Question
How can I make this code faster? It runs slow because it's a loop but I have not been able to come up with a solution that does not use loops. Any ideas or comments are appreciated.

If we can assume every group of 1's in entsig is followed by at least one 1 in
exsig, then you could compute ent, ex and in like this:
def ent_in_ex(df):
entsig_mask = (df['entsig'].diff().shift(1) == 1)
exsig_mask = (df['exsig'].diff().shift(1) == 1)
df.loc[entsig_mask, 'ent'] = 1
df.loc[exsig_mask, 'ex'] = 1
df['in'] = df['ent'].shift(1).cumsum().subtract(df['ex'].cumsum(), fill_value=0)
return df
If we can make this assumption, then ent_in_ex is significantly faster:
In [5]: %timeit orig(df)
10 loops, best of 3: 185 ms per loop
In [6]: %timeit ent_in_ex(df)
100 loops, best of 3: 2.23 ms per loop
In [95]: orig(df).equals(ent_in_ex(df))
Out[95]: True
where orig is the original code:
def orig(df):
for i in df.index:
df['ent'][(df.entsig.shift(1)==1) & (df['ent'].shift(1) == 0) & (df['in'].shift(1) == 0)]=1
df['ex'][(df.exsig.shift(1)==1) & (df['in'].shift(1)==1)]=1
df['in'][(df.ent.shift(1)==1) | ((df['in'].shift(1)==1) & (df['ex']==0))]=1
for j in df.index:
df['ent'][df['in'] == 1]=0
df['in'][df['ex']==1]=0
df['ex'][df['ex'].shift(1)==1]=0
return df