I have a dataframe of samples, with a country column. The relative number of records in each country are:
d1.groupby("country").size()
country
Australia 21
Cambodia 58
China 280
India 133
Indonesia 195
Malaysia 138
Myanmar 51
Philippines 49
Singapore 1268
Taiwan 47
Thailand 273
Vietnam 288
How do I select, say, 100 random samples from each country, if that country has > 100 samples? (if the country has <= 100 samples, do nothing). Currently, I do this for, say, Singapore:
names_nonsg_ls = []
names_sg_ls = []
# if the country is not SG, add it to names_nonsg_ls.
# else, add it to names_sg_ls, which will be subsampled later.
for index, row in d0.iterrows():
if str(row["country"]) != "Singapore":
names_nonsg_ls.append(str(row["header"]))
else:
names_sg_ls.append(str(row["header"]))
# Select 100 random names from names_sg_ls
names_sg_ls = random.sample(names_sg_ls, 100)
# Form the list of names to retain
names_ls = names_nonsg_ls + names_sg_ls
# create new dataframe
d1 = d0.loc[d0["header"].isin(names_ls)]
But manually a new list for each country that has >100 names is just poor form, not to mention that I first have to manually pick out the countries with > 100 names.
You can group by country, then sample based on the group size:
d1.groupby("country", group_keys=False).apply(lambda g: g.sample(100) if len(g) > 100 else g)
Example:
df = pd.DataFrame({
'A': ['a','a','a','a','b','b','b','c','d'],
'B': list(range(9))
})
df.groupby('A', group_keys=False).apply(lambda g: g.sample(3) if len(g) > 3 else g)
# A B
#2 a 2
#0 a 0
#1 a 1
#4 b 4
#5 b 5
#6 b 6
#7 c 7
#8 d 8
Related
I found this generic code online.
import pandas as pd
import holoviews as hv
from holoviews import opts, dim
from bokeh.sampledata.les_mis import data
hv.extension('bokeh')
hv.output(size=200)
links = pd.DataFrame(data['links'])
print(links.head(3))
hv.Chord(links)
nodes = hv.Dataset(pd.DataFrame(data['nodes']), 'index')
nodes.data.head()
chord = hv.Chord((links, nodes)).select(value=(5, None))
chord.opts(
opts.Chord(cmap='Category20', edge_cmap='Category20', edge_color=dim('source').str(),
labels='name', node_color=dim('index').str()))
That makes this, which looks nice.
[![enter image description here][1]][1]
The sample data is sourced from here.
https://holoviews.org/reference/elements/bokeh/Chord.html
Apparently, 'links' is a pandas dataframe and 'nodes' is a holoviews dataset, and the type is like this.
<class 'pandas.core.frame.DataFrame'>
<class 'holoviews.core.data.Dataset'>
So, my question is this...how can I feed a dataframe into a Chord Diagram? Here is my sample dataframe. Also, I don't know how to incorporate the <class 'holoviews.core.data.Dataset'> into the mix.
I think your data does not match the requirements of this function. Let me explain why I think so?
The Chord-function expects at least on dataset (this can be a pandas DataFrame) with three columns, but all elements are numbers.
source target value
0 1 0 1
1 2 0 8
2 3 0 10
A second dataset is optional. This can take strings in the second columns to add labels for example.
index name group
0 0 a 0
1 1 b 0
2 2 c 0
Basic Example
Your given data looks like this.
Measure Country Value
0 Arrivals Greece 1590
1 Arrivals Spain 1455
2 Arrivals France 1345
3 Arrivals Iceland 1100
4 Arrivals Iceland 1850
5 Departures America 2100
6 Departures Ireland 1000
7 Departures America 950
8 Departures Ireland 1200
9 Departures Japan 1050
You can bring your date in the basic form, if you replace the strings in your DataFrame df by numbers like this:
_df = df.copy()
values = list(_df.Measure.unique())+list(_df.Country.unique())
d = {value: i for i, value in enumerate(values)}
def str2num(s):
return d[s]
_df.Measure = _df.Measure.apply(str2num)
_df.Country = _df.Country.apply(str2num)
>>> df
Measure Country Value
0 0 2 1590
1 0 3 1455
2 0 4 1345
3 0 5 1100
4 0 5 1850
5 1 6 2100
6 1 7 1000
7 1 6 950
8 1 7 1200
9 1 8 1050
Now your data matches the basic conditions and you can create a Chord diagram.
chord = hv.Chord(_df).select(value=(5, None))
chord.opts(
opts.Chord(cmap='Category20', edge_cmap='Category20',
edge_color=dim('Measure').str(),
labels='Country',
node_color=dim('index').str()))
As you can see, all the conection lines only have one of two colors. This is because in the Measure column are only two elements. Therefor I think, this is not what you want.
Modificated Example
Let's Modify your data a tiny bit:
_list = list(df.Country.values)
new_df = pd.DataFrame({'From':_list, 'To':_list[3:]+_list[:3], 'Value':df.Value})
>>> new_df
From To Value
0 Greece Iceland 1590
1 Spain Iceland 1455
2 France America 1345
3 Iceland Ireland 1100
4 Iceland America 1850
5 America Ireland 2100
6 Ireland Japan 1000
7 America Greece 950
8 Ireland Spain 1200
9 Japan France 1050
and:
node = pd.DataFrame()
for i, value in enumerate(df.Measure.unique()):
_list = list(df[df['Measure']==value].Country.unique())
node = pd.concat([node, pd.DataFrame({'Name':_list, 'Group':i})], ignore_index=True)
>>> node
Name Group
0 Greece 0
1 Spain 0
2 France 0
3 Iceland 0
4 America 1
5 Ireland 1
6 Japan 1
Now we have to replace the strings in new_df again and can call the Chord-function again.
values = list(df.Country.unique())
d = {value: i for i, value in enumerate(values)}
def str2num(s):
return d[s]
new_df.From = new_df.From.apply(str2num)
new_df.To = new_df.To.apply(str2num)
hv.Chord(new_df)
nodes = hv.Dataset(pd.DataFrame(node), 'index')
chord = hv.Chord((new_df, nodes)).select(value=(5, None))
chord.opts(
opts.Chord(cmap='Category20', edge_cmap='Category20', edge_color=dim('From').str(),
labels='Name', node_color=dim('index').str()
)
)
The are now two groups added to the HoverTool.
I have the following data frame:
import pandas as pd
pandas_df = pd.DataFrame([
["SEX", "Male"],
["SEX", "Female"],
["EXACT_AGE", None],
["Country", "Afghanistan"],
["Country", "Albania"]],
columns=['FullName', 'ResponseLabel'
])
Now what I need to do is to add sort order to this dataframe. Each new "FullName" would increment it by 100 and each consecutive "ResponseLabel" for a given "FullName" would increment it by 1 (for this specific "FullName"). So I basically create two different sort orders that I sum later on.
pandas_full_name_increment = pandas_df[['FullName']].drop_duplicates()
pandas_full_name_increment = pandas_full_name_increment.reset_index()
pandas_full_name_increment.index += 1
pandas_full_name_increment['SortOrderFullName'] = pandas_full_name_increment.index * 100
pandas_df['SortOrderResponseLabel'] = pandas_df.groupby(['FullName']).cumcount() + 1
pandas_df = pd.merge(pandas_df, pandas_full_name_increment, on = ['FullName'], how = 'left')
Result:
FullName ResponseLabel SortOrderResponseLabel index SortOrderFullName SortOrder
0 SEX Male 1 0 100 101
1 SEX Female 2 0 100 102
2 EXACT_AGE NULL 1 2 200 201
3 Country Afghanistan 1 3 300 301
4 Country Albania 2 3 300 302
The result that I get on my "SortOrder" column is correct but I wonder if there is some better approach pandas-wise?
Thank you!
The best way to do this would be to use ngroup and cumcount
name_group = pandas_df.groupby('FullName')
pandas_df['sort_order'] = (
name_group.ngroup(ascending=False).add(1).mul(100) +
name_group.cumcount().add(1)
)
Output
FullName ResponseLabel sort_order
0 SEX Male 101
1 SEX Female 102
2 EXACT_AGE None 201
3 Country Afghanistan 301
4 Country Albania 302
I have a dataframe as shown below
Token Label StartID EndID
0 Germany Country 0 2
1 Berlin Capital 6 9
2 Frankfurt City 15 18
3 four million Number 21 24
4 Sweden Country 26 27
5 United Kingdom Country 32 34
6 ten million Number 40 45
7 London Capital 50 55
I am trying to get row based on certain condition, i.e. associate the label Number to closest capital i.e. Berlin
3 four million Number 21 24 - > 1 Berlin Capital 6 9
or something like:
df[row3] -> df [row1]
A pseudo logic
First check, for the rows with label: Number then (assumption is that the city is always '2 rows' above or below) and has the label: Capital. But, label: 'capital' loc is always after the label: Country
What I have done until now,
columnsName =['Token', 'Label', 'StartID', 'EndID']
df = pd.read_csv('resources/testcsv.csv', index_col= 0, skip_blank_lines=True, header=0)
print(df)
key_number = 'Number'
df_with_number = (df[df['Label'].str.lower().str.contains(r"\b{}\b".format(key_number), regex=True, case=False)])
print(df_with_number)
key_capital = 'Capital'
df_with_capitals = (df[df['Label'].str.lower().str.contains(r"\b{}\b".format(key_capital), regex=True, case=False)])
print(df_with_capitals)
key_country = 'Country'
df_with_country = (df[df[1].str.lower().str.contains(r"\b{}\b".format(key_country), regex=True, case=False)])
print(df_with_country)
The logic is to compare the index's and then make possible relations
i.e.
df[row3] -> [ df [row1], df[row7]]
you could use merge_asof with the parameter direction=nearest for example:
df_nb_cap = pd.merge_asof(df_with_number.reset_index(),
df_with_capitals.reset_index(),
on='index',
suffixes=('_nb', '_cap'), direction='nearest')
print (df_nb_cap)
index Token_nb Label_nb StartID_nb EndID_nb Token_cap Label_cap \
0 3 four_million Number 21 24 Berlin Capital
1 6 ten_million Number 40 45 London Capital
StartID_cap EndID_cap
0 6 9
1 50 55
# adjusted sample data
s = """Token,Label,StartID,EndID
Germany,Country,0,2
Berlin,Capital,6,9
Frankfurt,City,15,18
four million,Number,21,24
Sweden,Country,26,27
United Kingdom,Country,32,34
ten million,Number,40,45
London,Capital,50,55
ten million,Number,40,45
ten million,Number,40,45"""
df = pd.read_csv(StringIO(s))
# create a mask for number where capital is 2 above or below
# and where country is three above number or one below number
mask = (df['Label'] == 'Number') & (((df['Label'].shift(2) == 'Capital') |
(df['Label'].shift(-2) == 'Capital')) &
(df['Label'].shift(3) == 'Country') |
(df['Label'].shift(-1) == 'Country'))
# create a mask for capital where number is 2 above or below
# and where country is one above capital
mask2 = (df['Label'] == 'Capital') & (((df['Label'].shift(2) == 'Number') |
(df['Label'].shift(-2) == 'Number')) &
(df['Label'].shift(1) == 'Country'))
# hstack your two masks and create a frame
new_df = pd.DataFrame(np.hstack([df[mask].to_numpy(), df[mask2].to_numpy()]))
print(new_df)
0 1 2 3 4 5 6 7
0 four million Number 21 24 Berlin Capital 6 9
I have a dataframe that needs a column added to it. That column needs to be a count of all the other rows in the table that meet a certain condition, that condition needs to take in input both from the "input" row and the "output" row.
For example, if it was a dataframe describing people, and I wanted to make a column that counted how many people were taller than the current row and lighter.
I'd want the height and weight of the row, as well as the height and weight of the other rows in a function, so I can do something like:
def example_function(height1, weight1, height2, weight2):
if height1 > height2 and weight1 < weight2:
return True
else:
return False
And it would just sum up all the True's and give that sum in the column.
Is something like this possible?
Thanks in advance for any ideas!
Edit: Sample input:
id name height weight country
0 Adam 70 180 USA
1 Bill 65 190 CANADA
2 Chris 71 150 GERMANY
3 Eric 72 210 USA
4 Fred 74 160 FRANCE
5 Gary 75 220 MEXICO
6 Henry 61 230 SPAIN
The result would need to be:
id name height weight country new_column
0 Adam 70 180 USA 1
1 Bill 65 190 CANADA 1
2 Chris 71 150 GERMANY 3
3 Eric 72 210 USA 1
4 Fred 74 160 FRANCE 4
5 Gary 75 220 MEXICO 1
6 Henry 61 230 SPAIN 0
I believe it will need to be some sort of function, as the actual logic I need to use is more complicated.
edit 2:fixed typo
You can add booleans, like this:
count = ((df.height1 > df.height2) & (df.weight1 < df.weight2)).sum()
EDIT:
I test it a bit and then change conditions with custom function:
def f(x):
#check boolean mask
#print ((df.height > x.height) & (df.weight < x.weight))
return ((df.height < x.height) & (df.weight > x.weight)).sum()
df['new_column'] = df.apply(f, axis=1)
print (df)
id name height weight country new_column
0 0 Adam 70 180 USA 2
1 1 Bill 65 190 CANADA 1
2 2 Chris 71 150 GERMANY 3
3 3 Eric 72 210 USA 1
4 4 Fred 74 160 FRANCE 4
5 5 Gary 75 220 MEXICO 1
6 6 Henry 61 230 SPAIN 0
Explanation:
For each row compare values and for count simply sum values True.
For example, if it was a dataframe describing people, and I wanted to make a column that counted how many people were taller than the current row and lighter.
As far as I understand, you want to assign to a new column something like
df['num_heigher_and_leighter'] = df.apply(lambda r: ((df.height > r.height) & (df.weight < r.weight)).sum(), axis=1)
However, your text description doesn't seem to match the outcome, which is:
0 2
1 3
2 0
3 1
4 0
5 0
6 6
dtype: int64
Edit
As in any other case, you can use a named function instead of a lambda:
df = ...
def foo(r):
return ((df.height > r.height) & (df.weight < r.weight)).sum()
df['num_heigher_and_leighter'] = df.apply(foo, axis=1)
I'm assuming you had a typo and want to compare heights with heights and weights with weights. If so, you could count the number of persons taller OR heavier like so:
>>> for i,height,weight in zip(df.index,df.height, df.weight):
... cnt = df.loc[((df.height>height) & (df.weight>weight)), 'height'].count()
... df.loc[i,'thing'] = cnt
...
>>> df
name height weight country thing
0 Adam 70 180 USA 2.0
1 Bill 65 190 CANADA 2.0
2 Chris 71 150 GERMANY 3.0
3 Eric 72 210 USA 1.0
4 Fred 74 160 FRANCE 1.0
5 Gary 75 220 MEXICO 0.0
6 Henry 61 230 SPAIN 0.0
Here for instance, no person is Heavier than Henry, and no person is taller than Gary. If that's not what you intended, it should be easy to modify the & above to a | instead or switching out the > to a <.
When you're more accustomed to Pandas, I suggest you use Ami Tavory excellent answer instead.
PS. For the love of god, use the Metric system for representing weight and height, and convert to whatever for presentation. These numbers are totally nonsensical for the world population at large. :)
I am adding a column "state" into an existing dataframe that does not share a common column with my other data frame. Therefore, I need to convert zipcodes into states (example, 00704 would be PR) to load into the dataframe that has the new column state.
reviewers = pd.read_csv('reviewers.txt',
sep='|',
header=None,
names=['user id','age','gender','occupation','zipcode'])
reviewers['state'] = ""
user id age gender occupation zipcode state
0 1 24 M technician 85711
1 2 53 F other 94043
zipcodes = pd.read_csv('zipcodes.txt',
usecols = [1,4],
converters={'Zipcode':str})
Zipcode State
0 00704 PR
1 00704 PR
2 00704 PR
3 00704 PR
4 00704 PR
zipcodes1 = zipcodes.set_index('Zipcode') ###Setting the index to zipcode
dfzip = zipcodes1
print(dfzip)
State
Zipcode
00704 PR
00704 PR
00704 PR
zips = (pd.Series(dfzip.values.tolist(), index = zipcodes1['State'].index))
states = []
for zipcode in reviewers['Zipcode']:
if re.search('[a-zA-Z]+', zipcode):
append.states['canada']
elif zipcode in zips.index:
append.states(zips['zipcode'])
else:
append.states('unkown')
I am not sure if my loop is correct either. I have to sort the zipcodes by U.S zipcode (numerical), Canada zip codes(alphabetical), and then other zip codes which we define as (unknown). Let me know if you need the data file.
Use:
#remove duplicates and create Series for mapping
zips = zipcodes.drop_duplicates().set_index('Zipcode')['State']
#get mask for canada zip codes
#if possible small letters change to [a-zA-Z]+
mask = reviewers['zipcode'].str.match('[A-Z]+')
#new column by mask
reviewers['state'] = np.where(mask, 'canada', reviewers['zipcode'].map(zips))
#NaNs are replaced
reviewers['state'] = reviewers['state'].fillna('unknown')
Loop version with apply:
import re
def f(code):
res="unknown"
#if possible small letter change to [a-zA-Z]+
if re.match('[A-Z]+', code):
res='canada'
elif code in zips.index:
res=zips[code]
return res
reviewers['State1'] = reviewers['zipcode'].apply(f)
print (reviewers.tail(10))
user id age gender occupation zipcode state State1
933 934 61 M engineer 22902 VA VA
934 935 42 M doctor 66221 KS KS
935 936 24 M other 32789 FL FL
936 937 48 M educator 98072 WA WA
937 938 38 F technician 55038 MN MN
938 939 26 F student 33319 FL FL
939 940 32 M administrator 02215 MA MA
940 941 20 M student 97229 OR OR
941 942 48 F librarian 78209 TX TX
942 943 22 M student 77841 TX TX
#test if same output
print ((reviewers['State1'] == reviewers['state']).all())
True
Timings:
In [56]: %%timeit
...: mask = reviewers['zipcode'].str.match('[A-Z]+')
...: reviewers['state'] = np.where(mask, 'canada', reviewers['zipcode'].map(zips))
...: reviewers['state'] = reviewers['state'].fillna('unknown')
...:
100 loops, best of 3: 2.08 ms per loop
In [57]: %%timeit
...: reviewers['State1'] = reviewers['zipcode'].apply(f)
...:
100 loops, best of 3: 17 ms per loop
Your loop needs to be fixed:
states = []
for zipcode in reviewers['Zipcode']:
if re.match(r'\w+', zipcode):
states.extend('Canada')
elif zipcode in zips.index:
states.extend(zips[zipcode])
else:
states.extend('Unknown')
Also, am assuming you want the states list to be plugged back into the dataframe. In that case you don't need the for loop. You can use pandas apply on the dataframe to get a new column:
def findState(code):
res='Unknown'
if re.match(r'\w+', code):
res='Canada'
elif code in zips.index:
res=zips[code]
return res
reviewers['State'] = reviewers['Zipcode'].apply(findstate)