I have a dataframe of patients and their gene expressions. I has this format:
Patient_ID | gene1 | gene2 | ... | gene10000
p1 0.142 0.233 ... bla
p2 0.243 0.243 ... -0.364
...
p4000 1.423 bla ... -1.222
As you see, that dataframe contains noise, with cells that are values other then a float value.
I want to remove every row that has a any column with non numeric values.
I've managed to do this using apply and pd.to_numeric like this:
cols = df.columns[1:]
df[cols] = df[cols].apply(pd.to_numeric, errors='coerce')
df = df.dropna()
The problem is that it's taking for ever to run, and I need a better and more efficient way of achieving this
EDIT: To reproduce something like my data:
arr = np.random.random_sample((3000,10000))
df = pd.DataFrame(arr, columns=['gene' + str(i) for i in range(10000)])
df = pd.concat([pd.DataFrame(['p' + str(i) for i in range(10000)], columns=['Patient_ID']),df],axis = 1)
df['gene0'][2] = 'bla'
df['gene9998'][4] = 'bla'
Was right it is worth trying numpy :)
I got 30-60x times faster version (bigger array, larger improvement)
Convert to numpy array (.values)
Iterate through all rows
Try to convert each row to row of floats
If it fails (some NaN present), note this in boolean array
Create array based on the results
Code:
import pandas as pd
import numpy as np
from line_profiler_pycharm import profile
def op_version(df):
cols = df.columns[1:]
df[cols] = df[cols].apply(pd.to_numeric, errors='coerce')
return df.dropna()
def np_version(df):
keep = np.full(len(df), True)
for idx, row in enumerate(df.values[:, 1:]):
try:
row.astype(np.float)
except:
keep[idx] = False
pass # maybe its better to store to_remove list, depends on data
return df[keep]
#profile
def main():
arr = np.random.random_sample((3000, 5000))
df = pd.DataFrame(arr, columns=['gene' + str(i) for i in range(5000)])
df = pd.concat([pd.DataFrame(['p' + str(i) for i in range(3000)],
columns=['Patient_ID']), df], axis=1)
df['gene0'][2] = 'bla'
df['gene998'][4] = 'bla'
df2 = df.copy()
df = op_version(df)
df2 = np_version(df2)
Note I decreased number of columns so it is more feasible for tests.
Also, fixed small bug in your example, instead of:
df = pd.concat([pd.DataFrame(['p' + str(i) for i in range(10000)], columns=['Patient_ID']),df],axis = 1)
I think should be
df = pd.concat([pd.DataFrame(['p' + str(i) for i in range(3000)], columns=['Patient_ID']),df],axis = 1)
Related
I'm trying to achieve this kind of transformation with Pandas.
I made this code but unfortunately it doesn't give the result I'm searching for.
CODE :
import pandas as pd
df = pd.read_csv('file.csv', delimiter=';')
df = df.count().reset_index().T.reset_index()
df.columns = df.iloc[0]
df = df[1:]
df
RESULT :
Do you have any proposition ? Any help will be appreciated.
First create columns for test nonOK and then use named aggregatoin for count, sum column Values and for count Trues values use sum again, last sum both columns:
df = (df.assign(NumberOfTest1 = df['Test one'].eq('nonOK'),
NumberOfTest2 = df['Test two'].eq('nonOK'))
.groupby('Category', as_index=False)
.agg(NumberOfID = ('ID','size'),
Values = ('Values','sum'),
NumberOfTest1 = ('NumberOfTest1','sum'),
NumberOfTest2 = ('NumberOfTest2','sum'))
.assign(TotalTest = lambda x: x['NumberOfTest1'] + x['NumberOfTest2']))
I have two dataframes df1 and df2. I need to iterate on df1 rows to get the timestamp and based on this timestamp +- 1000 milliseconds I filter data from df2. The below code snippet explains it clearly,
dataframes = []
for i in df1.index:
tempdf = df1[df1.index.values == i]
attributeName = tempdf['EventId'].iloc[0]
timeStamp = tempdf['Timestamp'].iloc[0]
fehlerStatus = tempdf['FehlerStatus'].iloc[0]
tempdf2 = df2[(df2['DateTime']>=timeStamp + datetime.timedelta(milliseconds=0)) &
(df2['DateTime']<=timeStamp + datetime.timedelta(milliseconds=1000))].sort_values(by='DateTime', ascending=True).reset_index(drop=True)
if not tempdf2.empty:
tempdf2['TargetAttributePlusStatus'] = attributeName.replace(' ', '_') + fehlerStatus.replace(' ', '_')
dataframes.append(tempdf2)
df = pd.concat(dataframes, axis=0)
The code above takes forever to get executed. Is there a more convenient way to vectorize this?
EDITED
I want to write an If loop with conditions on cooncatenating strings.
i.e. If cell A1 contains a specific format of text, then only do you concatenate, else leave as is.
example:
If bill number looks like: CM2/0000/, then concatenate this string with the date column (month - year), else leave the bill number as it is.
Sample Data
You can create function which does what you need and use df.apply() to execute it on all rows.
I use example data from #Boomer answer.
EDIT: you didn't show what you really have in dataframe and it seems you have datetime in bill_date but I used strings. I had to convert strings to datetime to show how to work with this. And now it needs .strftime('%m-%y') or sometimes .dt.strftime('%m-%y') instead of .str[3:].str.replace('/','-'). Because pandas uses different formats to display dateitm for different countries so I couldn't use str(x) for this because it gives me 2019-09-15 00:00:00 instead of yours 15/09/19
import pandas as pd
df = pd.DataFrame({
'bill_number': ['CM2/0000/', 'CM2/0000', 'CM3/0000/', 'CM3/0000'],
'bill_date': ['15/09/19', '15/09/19', '15/09/19', '15/09/19']
})
df['bill_date'] = pd.to_datetime(df['bill_date'])
def convert(row):
if row['bill_number'].endswith('/'):
#return row['bill_number'] + row['bill_date'].str[3:].replace('/','-')
return row['bill_number'] + row['bill_date'].strftime('%m-%y')
else:
return row['bill_number']
df['bill_number'] = df.apply(convert, axis=1)
print(df)
Result:
bill_number bill_date
0 CM2/0000/09-19 15/09/19
1 CM2/0000 15/09/19
2 CM3/0000/09-19 15/09/19
3 CM3/0000 15/09/19
Second idea is to create mask
mask = df['bill_number'].str.endswith('/')
and later use it for all values
#df.loc[mask,'bill_number'] = df[mask]['bill_number'] + df[mask]['bill_date'].str[3:].str.replace('/','-')
df.loc[mask,'bill_number'] = df[mask]['bill_number'] + df[mask]['bill_date'].dt.strftime('%m-%y')
or
#df.loc[mask,'bill_number'] = df.loc[mask,'bill_number'] + df.loc[mask,'bill_date'].str[3:].str.replace('/','-')
df.loc[mask,'bill_number'] = df.loc[mask,'bill_number'] + df.loc[mask,'bill_date'].dt.strftime('%m-%y')
Left side needs .loc[mask,'bill_number'] instead of `[mask]['bill_number'] to correctly assing values - but right side doesn't need it.
import pandas as pd
df = pd.DataFrame({
'bill_number': ['CM2/0000/', 'CM2/0000', 'CM3/0000/', 'CM3/0000'],
'bill_date': ['15/09/19', '15/09/19', '15/09/19', '15/09/19']
})
df['bill_date'] = pd.to_datetime(df['bill_date'])
mask = df['bill_number'].str.endswith('/')
#df.loc[mask,'bill_number'] = df[mask]['bill_number'] + df[mask]['bill_date'].str[3:].str.replace('/','-')
# or
#df.loc[mask,'bill_number'] = df.loc[mask,'bill_number'] + df.loc[mask,'bill_date'].str[3:].str.replace('/','-')
df.loc[mask,'bill_number'] = df[mask]['bill_number'] + df[mask]['bill_date'].dt.strftime('%m-%y')
#or
#df.loc[mask,'bill_number'] = df.loc[mask,'bill_number'] + df.loc[mask,'bill_date'].dt.strftime('%m-%y')
print(df)
Third idea is to use numpy.where()
import pandas as pd
import numpy as np
df = pd.DataFrame({
'bill_number': ['CM2/0000/', 'CM2/0000', 'CM3/0000/', 'CM3/0000'],
'bill_date': ['15/09/19', '15/09/19', '15/09/19', '15/09/19']
})
df['bill_date'] = pd.to_datetime(df['bill_date'])
df['bill_number'] = np.where(
df['bill_number'].str.endswith('/'),
#df['bill_number'] + df['bill_date'].str[3:].str.replace('/','-'),
df['bill_number'] + df['bill_date'].dt.strftime('%m-%y'),
df['bill_number'])
print(df)
Maybe this will work for you. It would be nice to have a data sample like #Mike67 was stating. But based on your information this is what I came up with. Bulky, but it works. I'm sure someone else will have a fancier version.
import pandas as pd
from pandas import DataFrame, Series
dat = {'num': ['CM2/0000/','CM2/0000', 'CM3/0000/', 'CM3/0000',],
'date': ['15/09/19','15/09/19','15/09/19','15/09/19']}
df = pd.DataFrame(dat)
df['date'] = df['date'].map(lambda x: str(x)[3:])
df['date'] = df['date'].str.replace('/','-')
for cols in df.columns:
df.loc[df['num'].str.endswith('/'), cols] = df['num'] + df['date']
print(df)
Results:
num date
0 CM2/0000/09-19 09-19
1 CM2/0000 09-19
2 CM3/0000/09-19 09-19
3 CM3/0000 09-19
Every time I creat a loop function, it's common to have problem with the first one:
For example:
dfd = quandl.get("FRED/DEXBZUS")
dfe = quandl.get("ECB/EURBRL")
df = [dfd, dfe]
dps = []
for i in df:
I just get the second dataframe values.
Using this:
dfd = quandl.get("FRED/DEXBZUS")
df = [dfd]
dps = []
for i in df:
I got this:
Empty DataFrame
Columns: []
Index: []
And if I use this (repeting the first one):
dfd = quandl.get("FRED/DEXBZUS")
dfe = quandl.get("ECB/EURBRL")
df = [dfd, dfd, dfe]
dps = []
for i in df:
I get both dataframes correcly
Examples :
import quandl
import pandas as pd
#import matplotlib
import matplotlib.pyplot as plt
dfd = quandl.get("FRED/DEXBZUS")
dfe = quandl.get("ECB/EURBRL")
df = [dfd, dfe]
dps = []
for i in df:
df1 = i.reset_index()
results = pd.DataFrame(df1)
results = results.rename(columns={'Date': 'ds','Value': 'y'})
dps = pd.DataFrame(dps.append(results))
print(dps)
Empty DataFrame
Columns: []
Index: []
ds y
0 2008-01-02 2.6010
1 2008-01-03 2.5979
2 2008-01-04 2.5709
3 2008-01-07 2.6027
4 2008-01-08 2.5796
UPDATE
As Bruno suggested, it is related to this function:
dps = pd.DataFrame(dps.append(results))
How to append all the dataset into a one data frame ?
result=Pd.DataFrame(df1) If you create dataframe like this and don't give columns, then by default first it will take 1st row as column and later you are renaming columns what default created.
So please create pd.DataFrame(df1,columns=[column_list]).
First row will not skip.
#this will print every element in df
for i in df:
print i
Also,
for dfIndex, i in enumerate(df):
print i
print dfIndex #this will print the index of i in df
Note that indexes start at 0, not 1.
I have some trouble processing a big csv with Pandas. Csv consists of an index and about other 450 columns in groups of 3, something like this:
cola1 colb1 colc1 cola2 colb2 colc2 cola3 colb3 colc3
1 stra_1 ctrlb_1 retc_1 stra_1 ctrlb_1 retc_1 stra_1 ctrlb_1 retc_1
2 stra_2 ctrlb_2 retc_2 stra_2 ctrlb_2 retc_2 stra_2 ctrlb_2 retc_2
3 stra_3 ctrlb_3 retc_3 stra_3 ctrlb_3 retc_3 stra_3 ctrlb_3 retc_3
For each trio of columns I would like to "analyze B column (it's a sort of "CONTROL field" and depending on its value I should then return a value by processing col A and C.
Finally I need to return a concatenation of all resulting columns starting from 150 to 1.
I already tried with apply but it seems too slow (10 min to process 50k rows).
df['Path'] = df.apply(lambda x: getFullPath(x), axis=1)
with an example function you can find here:
https://pastebin.com/S9QWTGGV
I tried extracting a list of unique combinations of cola,colb,colc - preprocessing the list - and applying map to generate results and it speeds up a little:
for i in range(1,151):
df['Concat' + str(i)] = df['cola' + str(i)] + '|' + df['colb' + str(i)] + '|' + df['colc' + str(i)]
concats = []
for i in range(1,151):
concats.append('Concat' + str(i))
ret = df[concats].values.ravel()
uniq = list(set(ret))
list = {}
for member in ret:
list[member] = getPath2(member)
for i in range(1,MAX_COLS + 1):
df['Res' + str(i)] = df['Concat' + str(i)].map(list)
df['Path'] = df.apply(getFullPath2,axis=1)
function getPath and getFullPath2 are defined as example here:
https://pastebin.com/zpFF2wXD
But it seems still a little bit slow (6 min for processing everything)
Do you have any suggestion on how I could speed up csv processing?
I don't even know if the way I using to "concatenate" columns could be better :), tried with Series.cat but I didn't get how to chain only some columns and not the full df
Thanks very much!
Mic
Amended answer: I see from your criteria, you actually have multiple controls on each column. I think what works is to split these into 3 dataframes, applying your mapping as follows:
import pandas as pd
series = {
'cola1': pd.Series(['D_1','C_1','E_1'],index=[1,2,3]),
'colb1': pd.Series(['ret1','ret1','ret2'],index=[1,2,3]),
'colc1': pd.Series(['B_1','C_2','B_3'],index=[1,2,3]),
'cola2': pd.Series(['D_1','C_1','E_1'],index=[1,2,3]),
'colb2': pd.Series(['ret3','ret1','ret2'],index=[1,2,3]),
'colc2': pd.Series(['B_2','A_1','A_3'],index=[1,2,3]),
'cola3': pd.Series(['D_1','C_1','E_1'],index=[1,2,3]),
'colb3': pd.Series(['ret2','ret2','ret1'],index=[1,2,3]),
'colc3': pd.Series(['A_1','B_2','C_3'],index=[1,2,3]),
}
your_df = pd.DataFrame(series, index=[1,2,3], columns=['cola1','colb1','colc1','cola2','colb2','colc2','cola3','colb3','colc3'])
# Split your dataframe into three frames for each column type
bframes = your_df[[col for col in your_df.columns if 'colb' in col]]
aframes = your_df[[col for col in your_df.columns if 'cola' in col]]
cframes = your_df[[col for col in your_df.columns if 'colc' in col]]
for df in [bframes, aframes, cframes]:
df.columns = ['col1','col2','col3']
# Mapping criteria
def map_colb(c):
if c == 'ret1':
return 'A'
elif c == 'ret2':
return None
else:
return 'F'
def map_cola(a):
if a.startswith('D_'):
return 'D'
else:
return 'E'
def map_colc(c):
if c.startswith('B_'):
return 'B'
elif c.startswith('C_'):
return 'C'
elif c.startswith('A_'):
return None
else:
return 'F'
# Use it on each frame
aframes = aframes.applymap(map_cola)
bframes = bframes.applymap(map_colb)
cframes = cframes.applymap(map_colc)
# The trick here is filling 'None's from the left to right in order of precedence
final = bframes.fillna(cframes.fillna(aframes))
# Then just combine them using whatever delimiter you like
# df.values.tolist() turns a row into a list
pathlist = ['|'.join(item) for item in final.values.tolist()]
This gives a result of:
In[70]: pathlist
Out[71]: ['A|F|D', 'A|A|B', 'B|E|A']