firmai/pandasvault
Advanced Pandas Vault Utilities, Functions and Snippets.
| repo name | firmai/pandasvault |
| repo link | https://github.com/firmai/pandasvault |
| homepage | |
| language | Python |
| size (curr.) | 2977 kB |
| stars (curr.) | 160 |
| created | 2020-03-01 |
| license | |
PandasVault — Advanced Pandas Functions and Code Snippets
The only Pandas utility package you would ever need. It has no exotic external dependencies. All functions have been compared and tested with alternatives, only the fastest equivalent functions have been developed and included in this package. The package has more than 20 wrapped functions and 100 snippets.
Github PandasVault Link, LinkedIn
You have the option to view this Readme or run a Colab Notebook.
pip install pandasvault
If you can identify performance improvements, or improvements in code length and styling, please open a pull request. This package is new, all help and criticisms are appreciated. I would love to hear about any additional function ideas. If you have a function to contribute please open an issues tab or email me at d.snow(at)nyu.edu.
List of Code
Table Processing
- Configure Pandas
- Data Frame Formatting
- Data Frames for Testing
- Lower Case Columns
- Front and Back Column Selection
- Fast Data Frame Split
- Create Features and Labels List
- Short Basic Commands
- Read Commands
- Create Ordered Categories
- Select Columns Based on Regex
- Accessing Group of Groupby Object
- Multiple External Selection Criteria
- Memory Reduction Script
- Verify Primary Key
- Shift Columns to Front
- Multiple Column Assignment
- Method Changing Technique
- Load Multiple Files
- Drop Rows and Column Substring
- Explode a Column
- Nest List Back into Column
- Split Cells with List
Table Exploration
- Groupby Functionality
- Cross Correlation Series Without Duplicates
- Missing Data Report
- Duplicated Rows Report
- Skewness
Feature Processing
- Remove Correlated Pairs
- Replace Infrequently Occurring Categories
- Quasi-Constant Feature Detection
- Filling Missing Values Separately
- Conditioned Column Value Replacement
- Remove Non-numeric Values in Data Frame
- Feature Scaling, Normalisation, Standardisation
- Impute Null with Tail Distribution
- Detect Outliers
- Windzorise Outliers
- Drop Outliers
- Impute Outliers
Feature Engineering
- Automated Dummy Encoding
- Binarise Empty Columns
- Polynomials
- Transformations
- Genetic Programming
- Principal Component
- Multiple Lags
- Multiple Rolling
- Date Features
- Haversine Distance
- Parse Address
- Processing Strings in Pandas
- Filtering Strings in Pandas
Model Validation
List of Functions
import pandas as pd
import numpy as np
import pandasvault as pv
"""TABLE PROCESSING"""
df = pv.list_shuff(["target","c","d"],df)
df = pv.reduce_mem_usage(df)
"""TABLE EXPLORATION"""
df = pv.corr_list(df)
df = pv.missing_data(df)
"""FEATURE PROCESSING"""
df = pv.drop_corr(df, thresh=0.1,keep_cols=["target"])
df = pv.replace_small_cat(df,["cat"])
qconstant_col = pv.constant_feature_detect(data=df,threshold=0.9)
df_train, scl = pv.scaler(df,target="target",cols_ignore=["a"],type="MinMax")
df_test = pv.scaler(df_test,scaler=scl,train=False, target="target",cols_ignore=["a"])
df = pv.impute_null_with_tail(df,cols=df.columns)
index,para = pv.outlier_detect(df,"a",threshold=0.5,method="IQR")
df = pv.windsorization(data=df,col='a',para=para,strategy='both')
df = pv.impute_outlier(data=df,col='a', outlier_index=index,strategy='mean')
"""FEATURE EXTRACTION"""
df = pv.auto_dummy(df, unique=3)
df = pv.binarise_empty(df, frac=0.6)
df = pv.polynomials(df, ["a","b"])
df = pv.transformations(df,["a","b"])
df = pv.pca_feature(df,variance_or_components=0.80,drop_cols=["target","a"])
df = pv.multiple_lags(df, start=1, end=2,columns=["a","target"])
df = pv.multiple_rolling(df, columns=["a"])
df = pv.date_features(df, date="date_fake")
df['distance_central'] = df.apply(pv.haversine_distance,axis=1)
"""MODEL VALIDATION"""
scores = pv.classification_scores(y_test, y_predict, y_prob)
Functions and Snippets Applied
If you are running the code for the first time load this test dataframe:
!pip install pandasvault
import pandas as pd
import numpy as np
import pandasvault as pv
np.random.seed(1)
"""quick way to create a data frame for testing"""
df_test = pd.DataFrame(np.random.randn(3, 4), columns=['a', 'b', 'c', 'd']) \
.assign(target=lambda x: (x['b']+x['a']/x['d'])*x['c'])
Table Processing
»> Configure Pandas (func)
import pandas as pd
def pd_config():
options = {
'display': {
'max_colwidth': 25,
'expand_frame_repr': False, # Don't wrap to multiple pages
'max_rows': 14,
'max_seq_items': 50, # Max length of printed sequence
'precision': 4,
'show_dimensions': False
},
'mode': {
'chained_assignment': None # Controls SettingWithCopyWarning
}
}
for category, option in options.items():
for op, value in option.items():
pd.set_option(f'{category}.{op}', value) # Python 3.6+
if __name__ == '__main__':
pv.pd_config()
»> Data Frame Formatting
df = df_test.copy()
df["number"] = [3,10,1]
df_out = (
df.style.format({"a":"${:.2f}", "target":"${:.5f}"})
.hide_index()
.highlight_min("a", color ="red")
.highlight_max("a", color ="green")
.background_gradient(subset = "target", cmap ="Blues")
.bar("number", color = "lightblue", align = "zero")
.set_caption("DF with different stylings")
) ; df_out
See Colab for Output
»> Data Frames For Testing
df1 = pd.util.testing.makeDataFrame() # contains random values
print("Contains missing values")
df2 = pd.util.testing.makeMissingDataframe() # contains missing values
print("Contains datetime values")
df3 = pd.util.testing.makeTimeDataFrame() # contains datetime values
print("Contains mixed values")
df4 = pd.util.testing.makeMixedDataFrame(); df4.head() # contains mixed values
Contains missing values
Contains datetime values
Contains mixed values
»> Lower Case Columns
## Lower-case all DataFrame column names
df = df_test.copy() ; df
df.columns = ["A","BGs","c","dag","Target"]
df.columns = map(str.lower, df.columns); df
»> Front and Back Column Selection
df = df_test.copy(); df
def front(self, n):
return self.iloc[:, :n]
def back(self, n):
return self.iloc[:, -n:]
pd.back = back
pd.front = front
pd.back(df,2)
»> Fast Data Frame Split
df = df_test.copy(); df
test = df.sample(frac=0.4)
train = df[~df.isin(test)].dropna(); train
»> Create Features and Labels List
df = df_test.head()
y = 'target'
X = [name for name in df.columns if name not in [y, 'd']]
print('y =', y)
print('X =', X)
y = target
X = ['a', 'b', 'c']
»> Short Basic Commands
df = df_test.copy()
df["category"] = np.where( df["target"]>1, "1", "0")
df["k"] = df["category"].astype(str) +": " + df["d"].round(1).astype(str)
df = df.append(df, ignore_index=True) ; df.head()
"""set display width, col_width etc for interactive pandas session"""
pd.set_option('display.width', 200)
pd.set_option('display.max_colwidth', 20)
pd.set_option('display.max_rows', 100)
"""when you have an excel sheet with spaces in column names"""
df.columns = [c.lower().replace(' ', '_') for c in df.columns]
"""Add prefix to all columns"""
df.add_prefix("1_")
"""Add suffix to all columns"""
df.add_suffix("_Z")
"""Droping column where missing values are above a threshold"""
df.dropna(thresh = len(df)*0.95, axis = "columns")
"""Given a dataframe df to filter by a series ["a","b"]:"""
df[df['category'].isin(["1","0"])]
"""filter by multiple conditions in a dataframe df"""
df[(df['a'] >1) & (df['b'] <1)]
"""filter by conditions and the condition on row labels(index)"""
df[(df.a > 0) & (df.index.isin([0, 1]))]
"""regexp filters on strings (vectorized), use .* instead of *"""
df[df.category.str.contains(r'.*[0-9].*')]
"""logical NOT is like this"""
df[~df.category.str.contains(r'.*[0-9].*')]
"""creating complex filters using functions on rows"""
df[df.apply(lambda x: x['b'] > x['c'], axis=1)]
"""Pandas replace operation"""
df["a"].round(2).replace(0.87, 17, inplace=True)
df["a"][df["a"] < 4] = 19
"""Conditionals and selectors"""
df.loc[df["a"] > 1, ["a","b","target"]]
"""Selecting multiple column slices"""
df.iloc[:, np.r_[0:2, 4:5]]
"""apply and map examples"""
df[["a","b","c"]].applymap(lambda x: x+1)
"""add 2 to row 3 and return the series"""
df[["a","b","c"]].apply(lambda x: x[0]+2,axis=0)
"""add 3 to col A and return the series"""
df.apply(lambda x: x['a']+1,axis=1)
""" Split delimited values in a DataFrame column into two new columns """
df['new1'], df['new2'] = zip(*df['k'].apply(lambda x: x.split(': ', 1)))
""" Doing calculations with DataFrame columns that have missing values
In example below, swap in 0 for df['col1'] cells that contain null """
df['new3'] = np.where(pd.isnull(df['b']),0,df['a']) + df['c']
""" Exclude certain data type or include certain data types """
df.select_dtypes(exclude=['O','float'])
df.select_dtypes(include=['int'])
"""one liner to normalize a data frame"""
(df[["a","b"]] - df[["a","b"]].mean()) / (df[["a","b"]].max() - df[["a","b"]].min())
"""groupby used like a histogram to obtain counts on sub-ranges of a variable, pretty handy"""
df.groupby(pd.cut(df.a, range(0, 1, 2))).size()
"""use a local variable use inside a query of pandas using @"""
mean = df["a"].mean()
df.query("a > @mean")
"""Calculate the % of missing values in each column"""
df.isna().mean()
"""Calculate the % of missing values in each row"""
rows = df.isna().mean(axis=1) ; df.head()
»> Read Commands
df = pd.util.testing.makeMixedDataFrame()
df.to_csv("data.csv") ; df
"""To avoid Unnamed: 0 when loading a previously saved csv with index"""
"""To parse dates"""
"""To set data types"""
df_out = pd.read_csv("data.csv", index_col=0,
parse_dates=['D'],
dtype={"c":"category", "B":"int64"}).set_index("D")
"""Copy data to clipboard; like an excel copy and paste
df = pd.read_clipboard()
"""
"""Read table from website
df = pd.read_html(url, match="table_name")
"""
""" Read pdf into dataframe ()
!pip install tabula
from tabula import read_pdf
df = read_pdf('test.pdf', pages='all')
"""
df_out.head()
»> Create Ordered Categories
df = df_test.copy()
df["cats"] = ["bad","good","excellent"]; df
import pandas as pd
from pandas.api.types import CategoricalDtype
print("Let's create our own categorical order.")
cat_type = CategoricalDtype(["bad", "good", "excellent"], ordered = True)
df["cats"] = df["cats"].astype(cat_type)
print("Now we can use logical sorting.")
df = df.sort_values("cats", ascending = True)
print("We can also filter this as if they are numbers.")
df[df["cats"] > "bad"]
Let's create our own categorical order.
Now we can use logical sorting.
We can also filter this as if they are numbers.
»> Select Columns Based on Regex
df = df_test.head(); df
df.columns = ["a_l", "b_l", "c_r","d_r","target"] ; df
df_out = df.filter(regex="_l",axis=1) ; df_out
»> Accessing Group of Groupby Object
df = df_test.copy()
df = df.append(df, ignore_index=True)
df["groupie"] = ["falcon","hawk","hawk","eagle","falcon","hawk"]; df
gbdf = df.groupby("groupie")
hawk = gbdf.get_group("hawk").mean(); hawk
a 0.5012
b -0.9334
c 1.5563
d -1.6272
target -2.3938
dtype: float64
»> Multiple External Selection Criteria
df = df_test.copy()
cr1 = df["a"] > 0
cr2 = df["b"] < 0
cr3 = df["c"] > 0
cr4 = df["d"] >-1
df[cr1 & cr2 & cr3 & cr4]
»> Memory Reduction Script (func)
df = df_test.copy(); df
import gc
def reduce_mem_usage(df):
""" iterate through all the columns of a dataframe and modify the data type
to reduce memory usage.
"""
start_mem = df.memory_usage().sum() / 1024**2
print('Memory usage of dataframe is {:.2f} MB'.format(start_mem))
for col in df.columns:
col_type = df[col].dtype
gc.collect()
if col_type != object:
c_min = df[col].min()
c_max = df[col].max()
if str(col_type)[:3] == 'int':
if c_min > np.iinfo(np.int8).min and c_max < np.iinfo(np.int8).max:
df[col] = df[col].astype(np.int8)
elif c_min > np.iinfo(np.int16).min and c_max < np.iinfo(np.int16).max:
df[col] = df[col].astype(np.int16)
elif c_min > np.iinfo(np.int32).min and c_max < np.iinfo(np.int32).max:
df[col] = df[col].astype(np.int32)
elif c_min > np.iinfo(np.int64).min and c_max < np.iinfo(np.int64).max:
df[col] = df[col].astype(np.int64)
else:
if c_min > np.finfo(np.float16).min and c_max < np.finfo(np.float16).max:
df[col] = df[col].astype(np.float16)
elif c_min > np.finfo(np.float32).min and c_max < np.finfo(np.float32).max:
df[col] = df[col].astype(np.float32)
else:
df[col] = df[col].astype(np.float64)
else:
df[col] = df[col].astype('category')
end_mem = df.memory_usage().sum() / 1024**2
print('Memory usage after optimization is: {:.2f} MB'.format(end_mem))
print('Decreased by {:.1f}%'.format(100 * (start_mem - end_mem) / start_mem))
return df
df_out = pv.reduce_mem_usage(df); df_out
Memory usage of dataframe is 0.00 MB
Memory usage after optimization is: 0.00 MB
Decreased by 36.3%
»> Verify Primary Key (func)
df = df_test.copy()
df["first_d"] = [0,1,2]
df["second_d"] = [4,1,9] ; df
def verify_primary_key(df, column_list):
'''Verify if columns in column list can be treat as primary key'''
return df.shape[0] == df.groupby(column_list).size().reset_index().shape[0]
verify_primary_key(df, ["first_d","second_d"])
True
»> Shift Columns to Front (func)
df = df_test.copy(); df
def list_shuff(items, df):
"Bring a list of columns to the front"
cols = list(df)
for i in range(len(items)):
cols.insert(i, cols.pop(cols.index(items[i])))
df = df.loc[:, cols]
df.reset_index(drop=True, inplace=True)
return df
df_out = pv.list_shuff(["target","c","d"],df); df_out
»> Multiple Column Assignments
df = df_test.copy(); df
df_out = (df.assign(stringed = df["a"].astype(str),
ounces = df["b"]*12,# this will allow yo set a title
galons = lambda df: df["a"]/128)
.query("b > -1")
.style.set_caption("Average consumption")) ; df_out
»> Method Chaining Technique
df = df_test.copy()
df[df>df.mean()] = None ; df
# with line continuation character
df_out = df.dropna(subset=["b","c"],how="all") \
.loc[df["a"]>0] \
.round(2) \
.groupby(["target","b"]).max() \
.unstack() \
.fillna(0) \
.rolling(1).sum() \
.reset_index() \
.stack() \
.ffill().bfill()
df_out
»> Load Multiple Files
import os
os.makedirs("folder",exist_ok=True,); df_test.to_csv("folder/first.csv",index=False) ; df_test.to_csv("folder/last.csv",index=False)
import glob
files = glob.glob('folder/*.csv')
dfs = [pd.read_csv(fp) for fp in files]
df_out = pd.concat(dfs)
df_out
»> Drop Rows with Column Substring
df = df_test.copy()
df["string_feature"] = ["1xZoo", "Safe7x", "bat4"]; df
substring = ["xZ","7z", "tab4"]
df_out = df[~df.string_feature.str.contains('|'.join(substring))]; df_out
»> Unnest (Explode) a Column
df = df_test.head()
df["g"] = [[str(a)+lista for a in range(4)] for lista in ["a","b","c"]]; df
df_out = df.explode("g"); df_out.iloc[:5,:]
»> Nest List Back into Column
### Run above example first
df = df_out.copy()
df_out['g'] = df_out.groupby(df_out.index)['g'].agg(list); df_out.head()
»> Split Cells With Lists
df = df_test.head()
df["g"] = [",".join([str(a)+lista for a in range(4)]) for lista in ["a","b","c"]]; df
df_out = df.assign(g = df["g"].str.split(",")).explode("g"); df_out.head()
Table Exploration
»> Groupby Functionality
df = df_test.head()
df["gr"] = [1, 1 , 0] ;df
In [34]: gb.<TAB> # noqa: E225, E999
gb.agg gb.boxplot gb.cummin gb.describe gb.filter
gb.get_group gb.height gb.last gb.median gb.ngroups
gb.plot gb.rank gb.std gb.transform gb.aggregate
gb.count gb.cumprod gb.dtype gb.first gb.nth
gb.groups gb.hist gb.max gb.min gb.gender
gb.prod gb.resample gb.sum gb.var gb.ohlc
gb.apply gb.cummax gb.cumsum gb.fillna
gb.head gb.indices gb.mean gb.name
gb.quantile gb.size gb.tail gb.weight
df_out = df.groupby('gr').agg([np.sum, np.mean, np.std]); df_out.iloc[:,:8]
»> Cross Correlation Series Without Duplicates (func)
df = df_test.copy(); df
def corr_list(df):
return (df.corr()
.unstack()
.sort_values(kind="quicksort",ascending=False)
.drop_duplicates().iloc[1:]); df_out
pv.corr_list(df)
b target 0.9215
a d 0.6605
target 0.3206
b a -0.0724
c d -0.1764
b -0.4545
target d -0.4994
c target -0.7647
b d -0.7967
a c -0.8555
dtype: float64
»> Missing Data Report (func)
df = df_test.copy()
df[df>df.mean()] = None ; df
def missing_data(data):
"Create a dataframe with a percentage and count of missing values"
total = data.isnull().sum().sort_values(ascending = False)
percent = (data.isnull().sum()/data.isnull().count()*100).sort_values(ascending = False)
return pd.concat([total, percent], axis=1, keys=['Total', 'Percent'])
pv.df_out = missing_data(df); df_out
»> Duplicated Rows Report
df = df_test.copy()
df["a"].iloc[2] = df["a"].iloc[1]
df["b"].iloc[2] = df["b"].iloc[1] ; df
# Get a report of all duplicate records in a dataframe, based on specific columns
df_out = df[df.duplicated(['a', 'b'], keep=False)] ; df_out
»> Skewness (func)
df = df_test.copy(); df
from scipy.stats import skew
def display_skewness(data):
'''show skewness information
Parameters
----------
data: pandas dataframe
Return
------
df: pandas dataframe
'''
numeric_cols = data.columns[data.dtypes != 'object'].tolist()
skew_value = []
for i in numeric_cols:
skew_value += [skew(data[i])]
df = pd.concat(
[pd.Series(numeric_cols), pd.Series(data.dtypes[data.dtypes != 'object'].apply(lambda x: str(x)).values)
, pd.Series(skew_value)], axis=1)
df.columns = ['var_name', 'col_type', 'skew_value']
return df
display_skewness(df)
Feature Processing
»> Remove Correlated Pairs (func)
df= df_test.copy(); df
def drop_corr(df, thresh=0.99,keep_cols=[]):
df_corr = df.corr().abs()
upper = df_corr.where(np.triu(np.ones(df_corr.shape), k=1).astype(np.bool))
to_remove = [column for column in upper.columns if any(upper[column] > thresh)] ## Change to 99% for selection
to_remove = [x for x in to_remove if x not in keep_cols]
df_corr = df_corr.drop(columns = to_remove)
return df.drop(to_remove,axis=1)
df_out = pv.drop_corr(df, thresh=0.1,keep_cols=["target"]); df_out
»> Replace Infrequently Occuring Categories
df = df_test.copy()
df = df.append([df]*2)
df["cat"] = ["bat","bat","rat","mat","mat","mat","mat","mat","mat"]; df
def replace_small_cat(df, columns, thresh=0.2, term="other"):
for col in columns:
# Step 1: count the frequencies
frequencies = df[col].value_counts(normalize = True)
# Step 2: establish your threshold and filter the smaller categories
small_categories = frequencies[frequencies < thresh].index
df[col] = df[col].replace(small_categories, "Other")
return df
df_out = pv.replace_small_cat(df,["cat"]); df_out.head()
»> Quasi-Constant Features Detection (func)
df = df_test.copy()
df["a"] = 3
def constant_feature_detect(data,threshold=0.98):
""" detect features that show the same value for the
majority/all of the observations (constant/quasi-constant features)
Parameters
----------
data : pd.Dataframe
threshold : threshold to identify the variable as constant
Returns
-------
list of variables names
"""
data_copy = data.copy(deep=True)
quasi_constant_feature = []
for feature in data_copy.columns:
predominant = (data_copy[feature].value_counts() / np.float(
len(data_copy))).sort_values(ascending=False).values[0]
if predominant >= threshold:
quasi_constant_feature.append(feature)
print(len(quasi_constant_feature),' variables are found to be almost constant')
return quasi_constant_feature
# the original dataset has no constant variable
qconstant_col = pv.constant_feature_detect(data=df,threshold=0.9)
df_out = df.drop(qconstant_col, axis=1) ; df_out
1 variables are found to be almost constant
### I will take care of outliers separately
»> Filling Missing Values Separately
df = df_test.copy()
df[df>df.mean()] = None ; df
# Clean up missing values in multiple DataFrame columns
dict_fill = {'a': 4,
'b': 3,
'c': 5,
'd': 9999,
'target': "False"}
df = df.fillna(dict_fill) ;df
»> Conditioned Column Value Replacement
df = df_test.copy(); df
# Set DataFrame column values based on other column values
df.loc[(df['a'] >1 ) & (df['c'] <0), ['target']] = np.nan ;df
»> Remove Non-numeric Values in Data Frame
df = df_test.copy().assign(target=lambda row: row["a"].round(4).astype(str)+"SC"+row["b"].round(4).astype(str))
df["a"] = "TI4560L" + df["a"].round(4).astype(str) ; df
df_out = df.replace('[^0-9]+', '', regex=True); df_out
»> Feature Scaling, Normalisation, Standardisation (func)
df= df_test.copy(); df
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import MinMaxScaler
def scaler(df,scaler=None,train=True, target=None, cols_ignore=None, type="Standard"):
if cols_ignore:
hold = df[cols_ignore].copy()
df = df.drop(cols_ignore,axis=1)
if target:
x = df.drop([target],axis=1).values #returns a numpy array
else:
x = df.values
if train:
if type=="Standard":
scal = StandardScaler()
elif type=="MinMax":
scal = MinMaxScaler()
scal.fit(x)
x_scaled = scal.transform(x)
else:
x_scaled = scaler.transform(x)
if target:
df_out = pd.DataFrame(x_scaled, index=df.index, columns=df.drop([target],axis=1).columns)
df_out[target]= df[target]
else:
df_out = pd.DataFrame(x_scaled, index=df.index, columns=df.columns)
df_out = pd.concat((hold,df_out),axis=1)
if train:
return df_out, scal
else:
return df_out
df_out_train, scl = pv.scaler(df,target="target",cols_ignore=["a"],type="MinMax")
df_out_test = pv.scaler(df_test,scaler=scl,train=False, target="target",cols_ignore=["a"]); df_out_test
»> Impute Null with Tail Distribution (func)
df = df_test.copy()
df[df>df.mean()] = None ; df
def impute_null_with_tail(df,cols=[]):
"""
replacing the NA by values that are at the far end of the distribution of that variable
calculated by mean + 3*std
"""
df = df.copy(deep=True)
for i in cols:
if df[i].isnull().sum()>0:
df[i] = df[i].fillna(df[i].mean()+3*df[i].std())
else:
warn("Column %s has no missing" % i)
return df
df_out = pv.impute_null_with_tail(df,cols=df.columns); df_out
»> Detect Outliers (func)
df = df_test.copy(); df
def outlier_detect(data,col,threshold=3,method="IQR"):
if method == "IQR":
IQR = data[col].quantile(0.75) - data[col].quantile(0.25)
Lower_fence = data[col].quantile(0.25) - (IQR * threshold)
Upper_fence = data[col].quantile(0.75) + (IQR * threshold)
if method == "STD":
Upper_fence = data[col].mean() + threshold * data[col].std()
Lower_fence = data[col].mean() - threshold * data[col].std()
if method == "OWN":
Upper_fence = data[col].mean() + threshold * data[col].std()
Lower_fence = data[col].mean() - threshold * data[col].std()
if method =="MAD":
median = data[col].median()
median_absolute_deviation = np.median([np.abs(y - median) for y in data[col]])
modified_z_scores = pd.Series([0.6745 * (y - median) / median_absolute_deviation for y in data[col]])
outlier_index = np.abs(modified_z_scores) > threshold
print('Num of outlier detected:',outlier_index.value_counts()[1])
print('Proportion of outlier detected',outlier_index.value_counts()[1]/len(outlier_index))
return outlier_index, (median_absolute_deviation, median_absolute_deviation)
para = (Upper_fence, Lower_fence)
tmp = pd.concat([data[col]>Upper_fence,data[col]<Lower_fence],axis=1)
outlier_index = tmp.any(axis=1)
print('Num of outlier detected:',outlier_index.value_counts()[1])
print('Proportion of outlier detected',outlier_index.value_counts()[1]/len(outlier_index))
return outlier_index, para
index,para = pv.outlier_detect(df,"a",threshold=0.5,method="IQR")
print('Upper bound:',para[0],'\nLower bound:',para[1])
Num of outlier detected: 1
Proportion of outlier detected 0.3333333333333333
Upper bound: 1.5712030633954956
Lower bound: 0.2658967957893529
»> Windsorize Outliers (func)
# RUN above example first
df = df_test.copy(); df
def windsorization(data,col,para,strategy='both'):
"""
top-coding & bottom coding (capping the maximum of a distribution at an arbitrarily set value,vice versa)
"""
data_copy = data.copy(deep=True)
if strategy == 'both':
data_copy.loc[data_copy[col]>para[0],col] = para[0]
data_copy.loc[data_copy[col]<para[1],col] = para[1]
elif strategy == 'top':
data_copy.loc[data_copy[col]>para[0],col] = para[0]
elif strategy == 'bottom':
data_copy.loc[data_copy[col]<para[1],col] = para[1]
return data_copy
df_out = pv.windsorization(data=df,col='a',para=para,strategy='both'); df_out
»> Drop Outliers
## run the top two examples
df = df_test.copy(); df
df_out = df[~index] ; df_out
»> Impute Outliers
def impute_outlier(data,col,outlier_index,strategy='mean'):
"""
impute outlier with mean/median/most frequent values of that variable.
"""
data_copy = data.copy(deep=True)
if strategy=='mean':
data_copy.loc[outlier_index,col] = data_copy[col].mean()
elif strategy=='median':
data_copy.loc[outlier_index,col] = data_copy[col].median()
elif strategy=='mode':
data_copy.loc[outlier_index,col] = data_copy[col].mode()[0]
return data_copy
df_out = pv.impute_outlier(data=df,col='a', outlier_index=index,strategy='mean'); df_out
Feature Engineering
»> Automated Dummy (one-hot) Encoding (func)
df = df_test.copy()
df["e"] = np.where(df["c"]> df["a"], 1, 2)
def auto_dummy(df, unique=15):
# Creating dummies for small object uniques
if len(df)<unique:
raise ValueError('unique is set higher than data lenght')
list_dummies =[]
for col in df.columns:
if (len(df[col].unique()) < unique):
list_dummies.append(col)
print(col)
df_edit = pd.get_dummies(df, columns = list_dummies) # Saves original dataframe
#df_edit = pd.concat([df[["year","qtr"]],df_edit],axis=1)
return df_edit
df_out = pv.auto_dummy(df, unique=3); df_out
»> Binarise Empty Columns (func)
df = df_test.copy()
df[df>df.mean()] = None ; df
def binarise_empty(df, frac=80):
# Binarise slightly empty columns
this =[]
for col in df.columns:
if df[col].dtype != "object":
is_null = df[col].isnull().astype(int).sum()
if (is_null/df.shape[0]) >frac: # if more than 70% is null binarise
print(col)
this.append(col)
df[col] = df[col].astype(float)
df[col] = df[col].apply(lambda x: 0 if (np.isnan(x)) else 1)
df = pd.get_dummies(df, columns = this)
return df
df_out = pv.binarise_empty(df, frac=0.6); df_out
b
c
d
target
»> Polynomials (func)
df = df_test.copy(); df
def polynomials(df, feature_list):
for feat in feature_list:
for feat_two in feature_list:
if feat==feat_two:
continue
else:
df[feat+"/"+feat_two] = df[feat]/(df[feat_two]-df[feat_two].min()) #zero division guard
df[feat+"X"+feat_two] = df[feat]*(df[feat_two])
return df
df_out = pv.polynomials(df, ["a","b"]) ; df_out
»> Transformations (func)
df = df_test.copy(); df
def transformations(df,features):
df_new = df[features]
df_new = df_new - df_new.min()
sqr_name = [str(fa)+"_POWER_2" for fa in df_new.columns]
log_p_name = [str(fa)+"_LOG_p_one_abs" for fa in df_new.columns]
rec_p_name = [str(fa)+"_RECIP_p_one" for fa in df_new.columns]
sqrt_name = [str(fa)+"_SQRT_p_one" for fa in df_new.columns]
df_sqr = pd.DataFrame(np.power(df_new.values, 2),columns=sqr_name, index=df.index)
df_log = pd.DataFrame(np.log(df_new.add(1).abs().values),columns=log_p_name, index=df.index)
df_rec = pd.DataFrame(np.reciprocal(df_new.add(1).values),columns=rec_p_name, index=df.index)
df_sqrt = pd.DataFrame(np.sqrt(df_new.abs().add(1).values),columns=sqrt_name, index=df.index)
dfs = [df, df_sqr, df_log, df_rec, df_sqrt]
df= pd.concat(dfs, axis=1)
return df
df_out = pv.transformations(df,["a","b"]); df_out.iloc[:,:8]
»> Genetic Programming
! pip install gplearn
Collecting gplearn
[?25l Downloading https://files.pythonhosted.org/packages/43/6b/ee38cd74b32ad5056603aabbef622f9691f19d0869574dfc610034f18662/gplearn-0.4.1-py3-none-any.whl (41kB)
[K |████████████████████████████████| 51kB 2.5MB/s
[?25hRequirement already satisfied: scikit-learn>=0.20.0 in /usr/local/lib/python3.6/dist-packages (from gplearn) (0.22.1)
Requirement already satisfied: joblib>=0.13.0 in /usr/local/lib/python3.6/dist-packages (from gplearn) (0.14.1)
Requirement already satisfied: numpy>=1.11.0 in /usr/local/lib/python3.6/dist-packages (from scikit-learn>=0.20.0->gplearn) (1.17.5)
Requirement already satisfied: scipy>=0.17.0 in /usr/local/lib/python3.6/dist-packages (from scikit-learn>=0.20.0->gplearn) (1.4.1)
Installing collected packages: gplearn
Successfully installed gplearn-0.4.1
df = df_test.copy(); df
from gplearn.genetic import SymbolicTransformer
function_set = ['add', 'sub', 'mul', 'div',
'sqrt', 'log', 'abs', 'neg', 'inv','tan']
gp = SymbolicTransformer(generations=800, population_size=200,
hall_of_fame=100, n_components=10,
function_set=function_set,
parsimony_coefficient=0.0005,
max_samples=0.9, verbose=1,
random_state=0, n_jobs=6)
gen_feats = gp.fit_transform(df.drop("target", axis=1), df["target"]); df.iloc[:,:8]
df_out = pd.concat((df,pd.DataFrame(gen_feats, columns=["gen_"+str(a) for a in range(gen_feats.shape[1])])),axis=1); df_out.iloc[:,:8]
| Population Average | Best Individual |
---- ------------------------- ------------------------------------------ ----------
Gen Length Fitness Length Fitness OOB Fitness Time Left
0 10.14 0.91 22 1 0 43.36m
»> Prinicipal Component Features (func)
df =df_test.copy(); df
from sklearn.decomposition import PCA, IncrementalPCA
def pca_feature(df, memory_issues=False,mem_iss_component=False,variance_or_components=0.80,drop_cols=None):
if memory_issues:
if not mem_iss_component:
raise ValueError("If you have memory issues, you have to preselect mem_iss_component")
pca = IncrementalPCA(mem_iss_component)
else:
if variance_or_components>1:
pca = PCA(n_components=variance_or_components)
else: # automted selection based on variance
pca = PCA(n_components=variance_or_components,svd_solver="full")
X_pca = pca.fit_transform(df.drop(drop_cols,axis=1))
df = pd.concat((df[drop_cols],pd.DataFrame(X_pca, columns=["PCA_"+str(i+1) for i in range(X_pca.shape[1])])),axis=1)
return df
df_out = pv.pca_feature(df,variance_or_components=0.80,drop_cols=["target","a"]); df_out
»> Multiple Lags (func)
df = df_test.copy(); df
def multiple_lags(df, start=1, end=3,columns=None):
if not columns:
columns = df.columns.to_list()
lags = range(start, end+1) # Just two lags for demonstration.
df = df.assign(**{
'{}_t_{}'.format(col, t): df[col].shift(t)
for t in lags
for col in columns
})
return df
df_out = pv.multiple_lags(df, start=1, end=2,columns=["a","target"]); df_out
»> Multiple Rolling (func)
df = df_test.copy(); df
def multiple_rolling(df, windows = [1,2], functions=["mean","std"], columns=None):
windows = [1+a for a in windows]
if not columns:
columns = df.columns.to_list()
rolling_dfs = (df[columns].rolling(i) # 1. Create window
.agg(functions) # 1. Aggregate
.rename({col: '{0}_{1:d}'.format(col, i)
for col in columns}, axis=1) # 2. Rename columns
for i in windows) # For each window
df_out = pd.concat((df, *rolling_dfs), axis=1)
da = df_out.iloc[:,len(df.columns):]
da = [col[0] + "_" + col[1] for col in da.columns.to_list()]
df_out.columns = df.columns.to_list() + da
return df_out # 3. Concatenate dataframes
df_out = pv.multiple_rolling(df, columns=["a"]); df_out
»> Date Features
df = df_test.copy()
df["date_fake"] = pd.date_range(start="2019-01-03", end="2019-01-06", periods=len(df)); df
def date_features(df, date="date"):
df[date] = pd.to_datetime(df[date])
df[date+"_month"] = df[date].dt.month.astype(int)
df[date+"_year"] = df[date].dt.year.astype(int)
df[date+"_week"] = df[date].dt.week.astype(int)
df[date+"_day"] = df[date].dt.day.astype(int)
df[date+"_dayofweek"]= df[date].dt.dayofweek.astype(int)
df[date+"_dayofyear"]= df[date].dt.dayofyear.astype(int)
df[date+"_hour"] = df[date].dt.hour.astype(int)
df[date+"_int"] = pd.to_datetime(df[date]).astype(int)
return df
df_out = date_features(df, date="date_fake"); df_out.iloc[:,:8]
»> Haversine Distance (Location Feature) (func)
df = df_test.copy()
df["latitude"] = [39, 35 , 20]
df["longitude"]= [-77, -40 , -10 ]
from math import sin, cos, sqrt, atan2, radians
def haversine_distance(row, lon="latitude", lat="longitude"):
c_lat,c_long = radians(52.5200), radians(13.4050)
R = 6373.0
long = radians(row['longitude'])
lat = radians(row['latitude'])
dlon = long - c_long
dlat = lat - c_lat
a = sin(dlat / 2)**2 + cos(lat) * cos(c_lat) * sin(dlon / 2)**2
c = 2 * atan2(sqrt(a), sqrt(1 - a))
return R * c
df['distance_central'] = df.apply(pv.haversine_distance,axis=1); df.iloc[:,4:]
»> Parse Address
df = df_test.copy()
df["addr"] = pd.Series([
'Washington, D.C. 20003',
'Brooklyn, NY 11211-1755',
'Omaha, NE 68154' ]) ; df
regex = (r'(?P<city>[A-Za-z ]+), (?P<state>[A-Z]{2}) (?P<zip>\d{5}(?:-\d{4})?)')
df.addr.str.replace('.', '').str.extract(regex)
»> Processing Strings in Pandas
df = pd.util.testing.makeMixedDataFrame()
df["C"] = df["C"] + " " + df["C"] ; df
"""convert column to UPPERCASE"""
col_name = "C"
df[col_name].str.upper()
"""count string occurence in each row"""
df[col_name].str.count(r'\d') # counts number of digits
"""count # o chars in each row"""
df[col_name].str.count('o') # counts number of digits
"""split rows"""
s = pd.Series(["this is a regular sentence", "https://docs.p.org", np.nan])
s.str.split()
"""this creates new columns with the different split values (instead of lists)"""
s.str.split(expand=True)
"""limit the number of splits to 1, and start spliting from the rights side"""
s.str.rsplit("/", n=1, expand=True)
»> Filtering Strings in Pandas
df = pd.util.testing.makeMixedDataFrame()
df["C"] = df["C"] + " " + df["C"] ; df
col_name = "C"
"""check if a certain word/pattern occurs in each row"""
df[col_name].str.contains('oo') # returns True/False for each row
"""find occurences"""
df[col_name].str.findall(r'[ABC]\d') # returns a list of the found occurences of the specified pattern for each row
"""replace Weekdays by abbrevations (e.g. Monday --> Mon)"""
df[col_name].str.replace(r'(\w+day\b)', lambda x: x.groups[0][:3]) # () in r'' creates a group with one element, which we acces with x.groups[0]
"""create dataframe from regex groups (str.extract() uses first match of the pattern only)"""
df[col_name].str.extract(r'(\d?\d):(\d\d)')
df[col_name].str.extract(r'(?P<hours>\d?\d):(?P<minutes>\d\d)')
df[col_name].str.extract(r'(?P<time>(?P<hours>\d?\d):(?P<minutes>\d\d))')
"""if you want to take into account ALL matches in a row (not only first one):"""
df[col_name].str.extractall(r'(\d?\d):(\d\d)') # this generates a multiindex with level 1 = 'match', indicating the order of the match
df[col_name].replace('\n', '', regex=True, inplace=True)
"""remove all the characters after &# (including &#) for column - col_1"""
df[col_name].replace(' &#.*', '', regex=True, inplace=True)
"""remove white space at the beginning of string"""
df[col_name] = df[col_name].str.lstrip()
Model Validation
»> Classification Metrics (func)
y_test = [0, 1, 1, 1, 0]
y_predict = [0, 0, 1, 1, 1]
y_prob = [0.2,0.6,0.7,0.7,0.9]
from sklearn.metrics import roc_auc_score, average_precision_score, confusion_matrix
from sklearn.metrics import log_loss, brier_score_loss, accuracy_score
def classification_scores(y_test, y_predict, y_prob):
confusion_mat = confusion_matrix(y_test,y_predict)
TN = confusion_mat[0][0]
FP = confusion_mat[0][1]
TP = confusion_mat[1][1]
FN = confusion_mat[1][0]
TPR = TP/(TP+FN)
# Specificity or true negative rate
TNR = TN/(TN+FP)
# Precision or positive predictive value
PPV = TP/(TP+FP)
# Negative predictive value
NPV = TN/(TN+FN)
# Fall out or false positive rate
FPR = FP/(FP+TN)
# False negative rate
FNR = FN/(TP+FN)
# False discovery rate
FDR = FP/(TP+FP)
ll = log_loss(y_test, y_prob) # Its low but means nothing to me.
br = brier_score_loss(y_test, y_prob) # Its low but means nothing to me.
acc = accuracy_score(y_test, y_predict)
print(acc)
auc = roc_auc_score(y_test, y_prob)
print(auc)
prc = average_precision_score(y_test, y_prob)
data = np.array([np.arange(1)]*1).T
df_exec = pd.DataFrame(data)
df_exec["Average Log Likelihood"] = ll
df_exec["Brier Score Loss"] = br
df_exec["Accuracy Score"] = acc
df_exec["ROC AUC Sore"] = auc
df_exec["Average Precision Score"] = prc
df_exec["Precision - Bankrupt Firms"] = PPV
df_exec["False Positive Rate (p-value)"] = FPR
df_exec["Precision - Healthy Firms"] = NPV
df_exec["False Negative Rate (recall error)"] = FNR
df_exec["False Discovery Rate "] = FDR
df_exec["All Observations"] = TN + TP + FN + FP
df_exec["Bankruptcy Sample"] = TP + FN
df_exec["Healthy Sample"] = TN + FP
df_exec["Recalled Bankruptcy"] = TP + FP
df_exec["Correct (True Positives)"] = TP
df_exec["Incorrect (False Positives)"] = FP
df_exec["Recalled Healthy"] = TN + FN
df_exec["Correct (True Negatives)"] = TN
df_exec["Incorrect (False Negatives)"] = FN
df_exec = df_exec.T[1:]
df_exec.columns = ["Metrics"]
return df_exec
met = pv.classification_scores(y_test, y_predict, y_prob); met
0.6
0.5
