import pandas as pd

df = pd.read_csv('stanford_msa.csv') df.shape

# Get a sense of the distribution of the years of the incidents collections.Counter(df['Year'])

df.columns

df.head(20)

# Extract year fromm date df['Year'] = pd.DatetimeIndex(df['Date']).year

# Drop fields not used in analysis df_select = df.drop(['CaseID','Title','Location','Date','Shooter Sex','Relationship to Incident Location','Targeted Victim/s - General','Latitude','Longitude','City','Number of Victim Fatalities','Number of Victims Injured','Description','Day of Week','Date - Detailed','Shooter Name','Shooter Age(s)','Type of Gun - Detailed','Type of Gun - General','Fate of Shooter at the scene','Shooter\'s Cause of Death','Targeted Victim/s - Detailed','Possible Motive - Detailed','Possible Motive - General','History of Mental Illness - Detailed','Shooter Race','Place Type','Data Source 1','Data Source 2','Data Source 3','Data Source 4','Data Source 5','Data Source 6','Data Source 7','Military Experience'], axis=1)

# Remaining fields df_select.columns

# Since state and year are the attributes joining the two datasets, we drop the rows with either one of them absent df_select = df_select.dropna(subset=['State', 'Year'])

# Move `Year` field to the beginning year = df_select['Year'] df_select.drop(labels=['Year'], axis=1,inplace = True) df_select.insert(1, 'Year', year)

# We only have provision data from 1991 to 2016, so we will drop the data for the other years df_select = df_select[(df_select['Year'] >= 1991) & (df_select['Year'] <= 2016)] # We are left with 274 rows, which is not too bad

# Data cleaning to remove unknown df_select.replace('Unknown',0).head(20)

# Mean shooter age is 31.7 years df_select['Average Shooter Age'].astype(float).mean()

# 70 cases of shooters with history of mental illness df_select[df_select['History of Mental Illness - General'] > 0].count()

df_select.sort_values(by=['State','Year']).head(20)

# Data cleaning df_select = df_select.replace('32\r+ Unknown', 32) df_select = df_select.replace('No', 0) df_select = df_select.replace('no', 0) df_select = df_select.replace('Yes', 1) df_select = df_select.replace('Unknown', '0') df_select = df_select.replace('unknown', 0)

# Convert age into a binary variable indicating whether shooter is 21 years of younger df_select['Shooter Age <= 21'] = df_select.apply(lambda row: 1 if float(row['Average Shooter Age']) <= 21 else 0, axis=1)

# Convert data type (simplier methods did not work) for i in range(243): df_select.loc[i, 'Total Number of Guns'] = int(df_select.loc[i, 'Total Number of Guns'])

# Convert data type df_select['Number of Automatic Guns'] = df_select['Number of Automatic Guns'].astype(int) df_select['Number of Semi-Automatic Guns'] = df_select['Number of Semi-Automatic Guns'].astype(int) df_select['Total Number of Guns'] = df_select['Total Number of Guns'].astype(int) df_select['School Related'] = df_select['Total Number of Guns'].astype(int)

df_states = df_select.groupby(['State', 'Year'])[['Total Number of Fatalities','Total Number of Victims','Total Number of Guns','Number of Automatic Guns','Number of Semi-Automatic Guns','School Related','History of Mental Illness - General','Shooter Age <= 21']].agg('sum').reset_index()

df_states['Number of (Semi-)Automatic Guns'] = df_states['Number of Semi-Automatic Guns'] + df_states['Number of Automatic Guns']

df_states.columns

df_states_median

laws = pd.read_csv('provisions_data.csv')

print(laws.shape)

laws['year'].unique()

laws.head(7)

laws.sort_values(by=['state','year']).head(7)

laws.year.unique()

laws = laws[laws['year'] <= 2016]

df_select = df_select[df_select['Average Shooter Age'] != 'Unknown']

df_select = df_select.reset_index()

df_select = df_select.drop(index=116, axis=1).reset_index()

laws_by_incident = pd.DataFrame() for index, row in df_select.iterrows(): state = row['State'] year = row['Year'] laws_by_incident = laws_by_incident.append(laws[(laws['state'] == state) & (laws['year'] == year)])

print(df_select.shape) print(laws_by_incident.shape)

laws_by_incident = laws_by_incident.drop(['state','year'],axis=1)

df_select = df_select.replace('Unknown',0) df_select = df_select.replace('Yes',1) df_select = df_select.replace('No',0) df_select = df_select.drop(['Class'], axis=1) incidents = df_select.drop(['State','Year'],axis=1) incidents = incidents.replace('32\r+ Unknown', 32) incidents = incidents.replace('unknown', 0)

incidents

laws_by_incident

laws_by_state = pd.DataFrame() for index, row in df_states.iterrows(): state = row['State'] year = row['Year'] laws_by_state = laws_by_state.append(laws[(laws['state'] == state) & (laws['year'] == year)])

laws_by_state = laws_by_state.drop(['state','year'],axis=1)

states = df_states.drop(['State','Year','Number of Automatic Guns','Number of Semi-Automatic Guns'], axis=1)

states_median = df_states_median.drop(['Number of Automatic Guns','Number of Semi-Automatic Guns'], axis=1)

from sklearn.model_selection import train_test_split from sklearn.linear_model import LinearRegression from sklearn.linear_model import Ridge from sklearn.linear_model import Lasso, LassoCV from sklearn.kernel_ridge import KernelRidge from sklearn.linear_model import LogisticRegression from sklearn.feature_selection import SelectKBest, chi2 from sklearn.metrics import balanced_accuracy_score from sklearn import metrics

laws_by_incident.shape

y_age = incidents['Shooter Age <= 21']

y_mental = incidents['History of Mental Illness - General'] y_school = incidents['School Related']

def do_lin_regr(X, y): X_train,X_test,y_train,y_test = train_test_split(X, y,test_size=0.25,random_state=0) regr = LinearRegression() regr.fit(X_train,y_train) print(regr.score(X_test, y_test)) y_pred = regr.predict(X_test) print(balanced_accuracy_score(y_test, y_pred)) return regr.coef_

def do_ridge_regr(X, y, alpha=10): X_train,X_test,y_train,y_test = train_test_split(X, y,test_size=0.25,random_state=0) regr = Ridge(alpha=alpha) regr.fit(X_train,y_train) print(regr.score(X_test, y_test)) y_pred = regr.predict(X_test) print(balanced_accuracy_score(y_test, y_pred)) return regr.coef_

def do_lasso_regr(X, y, alpha=10): X_train,X_test,y_train,y_test = train_test_split(X, y,test_size=0.25,random_state=0) regr = Lasso(alpha=alpha) # regr = LassoCV() regr.fit(X_train,y_train) print(regr.score(X_test, y_test)) y_pred = regr.predict(X_test) print(balanced_accuracy_score(y_test, y_pred)) return regr.coef_

def do_log_regr(X, y): X_train,X_test,y_train,y_test = train_test_split(X, y,test_size=0.25,random_state=0) regr = LogisticRegression() regr.fit(X_train,y_train) print(regr.score(X_test, y_test)) y_pred = regr.predict(X_test) print(balanced_accuracy_score(y_test, y_pred)) return regr.coef_

import numpy.ma as ma

# Mental do_log_regr(laws_by_incident,y_mental)

y_school = y_school.replace('Killed',0) y_school = y_school.replace('no',0)

# School do_log_regr(laws_by_incident,y_school)

states.shape

median_cols = np.delete(df_states_median.columns, [3,4])

# Do regression for each column, i.e. statistic, in a for loop for col in median_cols: print(col) do_log_regr(laws_by_state, states_median[col])

coefs = do_lasso_regr(laws_by_state, states['Total Number of Fatalities'],5) mask = ma.masked_where(coefs > 0, coefs).mask laws_by_state.columns[mask]

def select_features(X, y, k=10): fs = SelectKBest(chi2, k=k) X_new = fs.fit_transform(X, y) feature_mask = fs.get_support() print(X.columns[feature_mask]) return X_new

select_features(laws_by_incident, y_school, 10)

X_age_new = select_features(laws_by_incident, y_age, 10) do_log_regr(X_age_new,y_age)

X_mental_new = select_features(laws_by_incident, y_mental) do_log_regr(X_mental_new,y_mental)

X_school_new = select_features(laws_by_incident, y_school, 15) do_log_regr(X_school_new,y_school)

for col in median_cols: print("********" + col) y = states_median[col] X_new = select_features(laws_by_state, y, 15) print(f'k=15') do_log_regr(X_new, y) X_new = select_features(laws_by_state, y, 10) print(f'k=10') do_log_regr(X_new, y)

top10 = ['assault', 'assaultlist', 'assaultregister', 'assaulttransfer', 'lockp', 'lockstandards', 'microstamp', 'onefeature', 'tenroundlimit', 'lawtotal','capuses', 'incidentall', 'loststolen', 'mcdvdating', 'opencarryh', 'personalized', 'preemptionnarrow', 'showing', 'waiting', 'lawtotal','ammrecords', 'ammrestrict', 'cap18', 'exparte', 'expartesurrender', 'mcdvremovalallowed', 'mcdvsurrenderdating', 'opencarryh', 'registration', 'lawtotal','ammlicense', 'danger', 'dvrosurrender', 'exparte', 'expartesurrender', 'invoutpatient', 'lockstandards', 'mcdvremovalallowed', 'mcdvsurrendernoconditions', 'lawtotal','ammlicense', 'danger', 'dvrosurrender', 'exparte', 'expartesurrender', 'invoutpatient', 'lockstandards', 'mcdvremovalallowed', 'mcdvsurrendernoconditions', 'lawtotal','ammrecords', 'dvro', 'dvrodating', 'dvrosurrenderdating', 'dvrosurrendernoconditions', 'expartedating', 'gunshow', 'mcdvsurrenderdating', 'universal', 'lawtotal','cap14', 'capuses', 'dvrosurrender', 'dvrosurrenderdating', 'dvrosurrendernoconditions', 'lockstandards', 'mcdvsurrender', 'mcdvsurrenderdating', 'mcdvsurrendernoconditions', 'lawtotal']

import collections top10_laws_count = collections.Counter(top10) top10_laws_count

# Handpicked FS X_age_manual = laws_by_incident[['age21handgunsale','age18longgunsale','age21longgunsaled','age21longgunsale','age21handgunpossess','age18longgunpossess','age21longgunpossess','lawtotal']] do_log_regr(X_age_manual,y_age)

# Automatic + handpicked FS X_age = laws_by_incident[['age21handgunsale','age18longgunsale','age21longgunsaled','age21longgunsale','age21handgunpossess','age18longgunpossess','age21longgunpossess','lawtotal', 'capunloaded', 'incidentall', 'inspection', 'liability', 'locked', 'loststolen', 'opencarryh', 'personalized','residential']] do_log_regr(X_age,y_age)

X_mental_manual = laws_by_incident[['lawtotal','invcommitment','invoutpatient','danger','drugmisdemeanor','alctreatment','alcoholism','universal','universalh']] do_log_regr(X_mental_manual,y_mental)

X_mental = laws_by_incident[['ammrecords', 'cap18', 'dealer', 'immunity', 'opencarryh', 'preemptionnarrow', 'registration', 'reportall', 'training', 'lawtotal','invcommitment','invoutpatient','danger','drugmisdemeanor','alctreatment','alcoholism','universal','universalh']]

do_log_regr(X_mental,y_mental)

X_school_manual = laws_by_incident[['college','collegeconcealed','elementary','opencarryh','opencarryl','opencarrypermith','opencarrypermitl','lawtotal']] do_log_regr(X_school_manual,y_school)

X_school = laws_by_incident[['college','collegeconcealed','elementary','opencarryh','opencarryl','opencarrypermith','opencarrypermitl','cap14', 'cap16', 'capaccess', 'capuses', 'collegeconcealed', 'dvro', 'dvrodating', 'dvrosurrendernoconditions', 'expartedating', 'liability', 'magazinepreowned', 'mcdv', 'nosyg', 'opencarrypermith', 'lawtotal']] do_log_regr(X_school,y_school)

X_age_new = select_features(X_age, y_age, 10) do_log_regr(X_age_new,y_age)

X_mental_new = select_features(X_mental, y_mental, 15) do_log_regr(X_mental_new,y_mental)

X_school_new = select_features(X_school, y_school, 15) do_log_regr(X_school_new,y_school)