Skip to content

Instantly share code, notes, and snippets.

@bolshoibooze
Forked from why-not/gist:4582705
Last active November 26, 2016 13:46
Show Gist options
  • Save bolshoibooze/2c0f9774241c81d2c671 to your computer and use it in GitHub Desktop.
Save bolshoibooze/2c0f9774241c81d2c671 to your computer and use it in GitHub Desktop.
Pandas recipe. I find pandas indexing counter intuitive, perhaps my intuitions were shaped by many years in the imperative world. I am collecting some recipes to do things quickly in pandas & to jog my memory.
"""quick way to create a data frame to try things out"""
df = pd.DataFrame(np.random.randn(5, 4), columns=['a', 'b', 'c', 'd'])
df['A'] """ will bring out a col """ df.ix[0] """will bring out a row, #0 in this case"""
"""to get an array from a data frame or a series use values, note it is not a function here, so no parans ()"""
point = df_allpoints[df_allpoints['names'] == given_point] # extract one point row.
point = point['desc'].values[0] # get its descriptor in array form.
"""Given a dataframe df to filter by a series s:"""
df[df['col_name'].isin(s)]
"""to do the same filter on the index instead of arbitrary column"""
df.ix[s]
""" display only certain columns, note it is a list inside the parans """
df[['A', 'B']]
"""drop rows with atleast one null value, pass params to modify
to atmost instead of atleast etc."""
df.dropna()
"""deleting a column"""
del df['column-name'] # note that df.column-name won't work.
"""making rows out of whole objects instead of parsing them into seperate columns"""
# Create the dataset (no data or just the indexes)
dataset = pandas.DataFrame(index=names)
# Add a column to the dataset where each column entry is a 1-D array and each row of “svd” is applied to a different DataFrame row
dataset['Norm']=svds
"""filter by multiple conditions in a dataframe df
parentheses!"""
df[(df['gender'] == 'M') & (df['cc_iso'] == 'US')]
"""filter by conditions and the condition on row labels(index)"""
df[(df.a > 0) & (df.index.isin([0, 2, 4]))]
"""regexp filters on strings (vectorized), use .* instead of *"""
df[df.category.str.contains(r'some.regex.*pattern')]
"""logical NOT is like this"""
df[~df.category.str.contains(r'some.regex.*pattern')]
"""creating complex filters using functions on rows: http://goo.gl/r57b1"""
df[df.apply(lambda x: x['b'] > x['c'], axis=1)]
"""Pandas replace operation http://goo.gl/DJphs"""
df[2].replace(4, 17, inplace=True)
df[1][df[1] == 4] = 19
"""apply and map examples"""
"""add 1 to every element"""
df.applymap(lambda x: x+1)
"""add 2 to row 3 and return the series"""
df.apply(lambda x: x[3]+2,axis=0)
"""add 3 to col A and return the series"""
df.apply(lambda x: x['a']+1,axis=1)
"""assigning some value to a slice is tricky as sometimes a copy is returned,
sometimes a view is returned based on numpy rules, more here:
http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-advanced"""
df.ix[df['part'].isin(ids), 'assigned_name'] = "some new value"
"""example of applying a complex external function
to each row of a data frame"""
def stripper(x):
l = re.findall(r'[0-9]+(?:\.[0-9]+){3}', x['Text with IP adress embedded'])
# you can take care of special
# cases and missing values, more than expected
# number of return values etc like this.
if l == []:
return ''
else:
return l[0]
df.apply(stripper, axis=1)
"""can pass extra args and named ones eg.."""
def subtract_and_divide(x, sub, divide=1):
return (x - sub) / divide
"""You may then apply this function as follows:"""
df.apply(subtract_and_divide, args=(5,), divide=3)
"""compute the means by group, and save mean to every element so group mean is available for every sample"""
sil_means = df.groupby('labels').mean()
df = df.join(sil_means, on='labels', rsuffix='_mean')
"""groupby used like a histogram to obtain counts on sub-ranges of a variable, pretty handy"""
df.groupby(pd.cut(df.age, range(0, 130, 10))).size()
"""finding the distribution based on quantiles"""
df.groupby(pd.qcut(df.age, [0, 0.99, 1])
"""if you don't need specific bins like above, and just want to count number of each values"""
df.age.value_counts()
"""one liner to normalize a data frame"""
(df - df.mean()) / (df.max() - df.min())
"""iterating and working with groups is easy when you realize each group is itself a DataFrame"""
for name, group in dg:
print name, print(type(group))
"""grouping and applying a group specific function to each group element,
I think this could be simpler, but here is my current version"""
quantile = [0, 0.50, 0.75, 0.90, 0.95, 0.99, 1]
grouped = df.groupby(pd.qcut(df.age, quantile))
frame_list = []
for i, group in enumerate(grouped):
(label, frame) = group
frame['age_quantile'] = quantile[i + 1]
frame_list.append(frame)
df = pd.concat(frame_list)
# more useful tricks::http://social-metrics.org/python-pandas-cookbook/
""" 1:Create list from dataframe column:"""
my_list = df[‘column_name’].tolist()
""" 2.Fill missing values of a column with zero """
df['column_name'].fillna(0)
""" 3. Membership of values:outlandish example """
df[(df['column_name']<= x) & (df.index.isin([0,1,2,3,4,5,6]))]
""" 4. Another way to use agrsort() """
df.ix[(df.column_name - x).abs().argsort() ]
# deep dive into pandas::http://chrisalbon.com/#Machine_Learning
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment