def make_ols_sklearn(X, y, test_size=0.20, fit_intercept=False, standardize=False):
"""
Makes an OLS in sklearn
"""
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)
if standardize:
ss = StandardScaler()
ss.fit(X_train)
X_train = ss.transform(X_train)
X_test = ss.transform(X_test)
ols = LinearRegression(fit_intercept=fit_intercept, normalize=False)
ols.fit(X_train, y_train)
train_score = ols.score(X_train, y_train)
test_score = ols.score(X_test, y_test)
cvmae_5 = np.mean(cross_val_score(ols,X , y, cv=5, scoring='neg_mean_absolute_error'))
cvmae_10 = np.mean(cross_val_score(ols,X, y, cv=10, scoring='neg_mean_absolute_error'))
print(f"train R2 score = {train_score}")
print(f"test R2 score = {test_score}\n")
print(f"cv5 MSE score = {cvmae_5}")
print(f"cv10 MSE score = {cvmae_10}")
return ols
def cleanRP_SG(self):
from math import exp
NoneOrZero=['BsmtQual','BsmtCond','BsmtExposure','BsmtFinType1',
'BsmtFinType2','BsmtFinSF1','BsmtFinSF2','Alley',
'Fence','GarageType','GarageQual',
'GarageCond','GarageFinish','GarageCars',
'GarageArea','MasVnrArea','MasVnrType','MiscFeature','PoolQC',
'BsmtFullBath', 'BsmtHalfBath', 'BsmtUnfSF']
mode=['Electrical','Exterior1st','Exterior2nd','FireplaceQu','Functional','KitchenQual','MSZoning','SaleType','Utilities']
mean=['TotalBsmtSF']
columns_with_missing_data=[name for name in self.all.columns if np.sum(self.all[name].isnull()) !=0]
columns_with_missing_data.remove('SalePrice')
for column in columns_with_missing_data:
col_data = self.all[column]
#print( 'Cleaning ' + str(np.sum(col_data.isnull())) + ' data entries for column: ' + column )
#log transformation for missing LotFrontage
if column=='LotFrontage':
ols=linear_model.LinearRegression()
my_ind=self.all[self.all['LotFrontage'].isnull()].index
y_train=self.all['LotFrontage'].loc[self.all['LotFrontage'].isnull()==False].values
x_train=self.all['LotArea'].loc[self.all['LotFrontage'].isnull()==False].values
x_train=np.log(x_train)
ols.fit(x_train.reshape(-1,1), y_train) #### What happen if we remove the 'reshape' method?
for i in my_ind:
print(np.log(self.all['LotArea'].loc[i])*ols.coef_)
self.all.loc[i,'LotFrontage']=((np.log(self.all.loc[i,'LotArea'])*ols.coef_)+ols.intercept_)[0]
#imputing the value of YearBuiltto the GarageYrBlt.
elif column=='GarageYrBlt':
missing_index=self.all[self.all['GarageYrBlt'].isnull()].index
for i in missing_index:
self.all.loc[i,'GarageYrBlt']=1871
elif column in mode:
# in case of function messing up - remove [0]
self.all[column] = [self.all[column].mode()[0] if pd.isnull(x) else x for x in self.all[column]]
elif column in mean:
self.all[column].fillna(self.all[column].mean(),inplace=True)
elif column in NoneOrZero:
if col_data.dtype == 'object':
no_string = 'None'
self.all[column] = [ no_string if pd.isnull(x) else x for x in self.all[column]]
else:
self.all[column] = [ 0 if pd.isnull(x) else x for x in self.all[column]]
else:
print( 'Uh oh!!! No cleaning strategy for:' + column )
def regression(df):
msk = np.random.rand(len(df)) < 0.75
train = df[msk]
test = df[~msk]
ols = sm.OLS(train['Number of total People injured and killed'], train.drop('Number of total People injured and killed', 1))
result = ols.fit()
return result.summary()
def regression(df):
msk = np.random.rand(len(df)) < 0.75
train = df[msk]
test = df[~msk]
ols = sm.OLS(train['Number of total People injured and killed'],
train.drop('Number of total People injured and killed', 1))
result = ols.fit()
return result.summary()
def _regression(self):
'''
generate a summary table about the parameters from the regression analysis
'''
mask = np.random.rand(len(self._data)) < 0.75
train = self._data[mask]
test = self._data[~mask]
ols = sm.OLS(train['TOTAL FATALITIES'],
train.drop('TOTAL FATALITIES', 1))
result = ols.fit()
return result.summary()
def regression(self):
'''
generate a summary table about the parameters from the regression analysis
'''
df = self._data
msk = np.random.rand(len(df)) < 0.75
train = df[msk]
test = df[~msk]
ols = sm.OLS(train['Total Fatalities'], train.drop('Total Fatalities', 1))
result = ols.fit()
print result.summary()
def sg_simpleLM(self):
self.test_train_split()
x = np.asarray(self.x_train)
# x = sm.add_constant(x)
ols = linear_model.LinearRegression()
# ols = sm.OLS(np.asarray(self.y_train),)
model = ols.fit(x, np.asarray(self.y_train)) #.reshape(-1,1)
sm_model_pred = model.predict(self.x_test)
self.plot_results(sm_model_pred)
print('RMSLE from Kaggle: ' +
str(self.rmsle(y_pred=sm_model_pred, y_test=self.y_test)))
print(self.rmse_cv(model, self.x_train, self.y_train))
self.results_dict['simpleLM'] = self.rmse_cv(model, self.x_train,
self.y_train)
def make_ols(df, x_columns, target='price'):
"""Pass in a DataFrame & your predictive columns to return an OLS regression model """
#set your x and y variables
X = df[x_columns]
y = df[target]
# pass them into stats models OLS package
ols = sm.OLS(y, X)
#fit your model
model = ols.fit()
#display the model summarry
display(model.summary())
#plot the residuals
fig = sm.graphics.qqplot(model.resid, dist=stats.norm, line='r', alpha=.65, fit=True, markerfacecolor="y")
plt.xlim(-2.5, 2.5)
plt.ylim(-2.5, 2.5)
#return model for later use
return
sc = StandardScaler() X_train_S=sc.fit_transform(X_train) X_test_S=sc.transform(X_test) model_lr= LinearRegression() model_lr.fit(X_train_S, y_train) y_pred_lr = model_lr.predict(X_test_S) print(model_lr.intercept_) print(model_lr.coef_) ols = sm.OLS(y_train, X_train_S) lr=ols.fit() pvalues = lr.pvalues print(lr.summary()) yF = fulldata_NMiss['Premium'] XF = fulldata_NMiss[['ins_Cancelation', 'Ticket', 'Accident']] XF_train, XF_test, yF_train, yF_test = train_test_split(XF, yF, test_size=0.25, random_state=100) sc = StandardScaler() XF_train_S=sc.fit_transform(XF_train) XF_test_S=sc.transform(XF_test) model_lr= LinearRegression() model_lr.fit(XF_train_S, yF_train)