import numpy as np
import matplotlib.pyplot as plt
import mglearn
import pandas as pd
import pylab as plot
import lmfit as lf
import random
import seaborn as sns
%matplotlib inline
from sklearn.linear_model import LogisticRegression
from sklearn.linear_model import LinearRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.neighbors import KNeighborsRegressor
from sklearn.model_selection import train_test_split
from sklearn.svm import LinearSVC
from sklearn.linear_model import Lasso
from sklearn.linear_model import Ridge
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import StandardScaler
import warnings
warnings.simplefilter(action='ignore')
pd.options.display.float_format = '{:,.2g}'.formatC:\Users\63917\Anaconda3\lib\site-packages\sklearn\externals\six.py:31: DeprecationWarning: The module is deprecated in version 0.21 and will be removed in version 0.23 since we've dropped support for Python 2.7. Please rely on the official version of six (https://pypi.org/project/six/).
"(https://pypi.org/project/six/).", DeprecationWarning)
C:\Users\63917\Anaconda3\lib\site-packages\sklearn\externals\joblib\__init__.py:15: DeprecationWarning: sklearn.externals.joblib is deprecated in 0.21 and will be removed in 0.23. Please import this functionality directly from joblib, which can be installed with: pip install joblib. If this warning is raised when loading pickled models, you may need to re-serialize those models with scikit-learn 0.21+.
warnings.warn(msg, category=DeprecationWarning)
Framework:
- EDA - look for inconsistencies wrt competition
- Total lost money for Outcome = 0
- Calculate Sum of (Amount * [(APR-Cost of Fund)/100] * (Term/12))
- Targeting & Segmentation (James)
Cluster by Loan Terms
- inconsistent outcome
- low outcome
- Maximize interest (best combination of term & APR to make outcome = 1) - ML Model (Elijah, Angela, Ria)
- Simulate declined to recalculate potential profit
Your task is to help Nomis Solutions to study and explore e-Car's data and to help them highlight potential inefficiencies in e-Car's current model and convince the company that there was value to be captured.
Guide on approaching the task:
- Perform exploratory data analysis.
- How will you show that the current pricing technique contains "pricing errors"?
- How do we know that there are "pricing errors"? What are we trying to maximize in this project?
- How can we make the analysis more manageable? Should we segment the customers?
- How should we price the loans?
- Do you think you can recommend the right price to quote?
- Has there been a mis-pricing of APR quotes?
- Can we build a systematic approach that can scale with the number of segments given certain customer characteristics?
df = pd.read_excel("NomisB.xlsx", na_values=' ')df_N = df[df['Car Type']=='N']df_U = df[df['Car Type']=='U']
df_U = df_U.fillna(0)
df_U = pd.get_dummies(df_U, columns=['Partner Bin'])
df_U = df_U.drop(columns=['Approve Date', 'Car Type', 'Previous Rate'])
df_U.head()
df_U_target = df_U['Outcome']
df_U_features = df_U.drop('Outcome', axis=1)df_R = df[df['Car Type']=='R']
df_R = df_R.fillna(0)
df_R = pd.get_dummies(df_R, columns=['Partner Bin'])
df_R = df_R.drop(columns=['Approve Date', 'Car Type'])
df_R.head()
df_R_target = df_R['Outcome']
df_R_features = df_R.drop('Outcome', axis=1)import random
df_a = df_N[df_N['Outcome']==0].reset_index()
data_pts = df_a.shape[0]
rand_pts = random.sample(range(data_pts), 9000)
df_a = df_a.iloc[rand_pts]
df_b = df_N[df_N['Outcome']==1].reset_index()
data_pts = df_b.shape[0]
rand_pts = random.sample(range(data_pts), 9000)
df_b = df_b.iloc[rand_pts]
df_b.head()
df_N = df_a.append(df_b).reset_index().drop(['level_0','index'], axis=1)df_N = df_N.fillna(0)
df_N.shape(119059, 12)
df_N = pd.get_dummies(df_N, columns=['Partner Bin'])df_N = df_N.drop(columns=['Approve Date', 'Car Type', 'Previous Rate'])df_N.head()| Tier | FICO | Term | Amount | Competition rate | Outcome | Rate | Cost of Funds | Partner Bin_1 | Partner Bin_2 | Partner Bin_3 | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 1 | 743 | 36 | 1.9e+04 | 5 | 1 | 4.8 | 1.8 | 1 | 0 | 0 |
| 1 | 1 | 752 | 60 | 3.6e+04 | 5.7 | 1 | 5.5 | 1.8 | 1 | 0 | 0 |
| 3 | 2 | 724 | 60 | 1.9e+04 | 5.7 | 1 | 5.4 | 1.8 | 0 | 0 | 1 |
| 4 | 2 | 700 | 72 | 2.4e+04 | 6.2 | 1 | 7 | 1.8 | 0 | 0 | 1 |
| 8 | 1 | 779 | 72 | 2e+04 | 6.2 | 1 | 6.6 | 1.8 | 1 | 0 | 0 |
df_N_target = df_N['Outcome']df_N_features = df_N.drop(columns='Outcome')import matplotlib.pyplot as plt
fig, axes = plt.subplots(nrows=2, ncols=7, figsize=(15, 10))
for i, column in enumerate(df_N.columns):
sns.distplot(df_N[column],ax=axes[i//7,i%7])
plt.suptitle('Distribution of Numeric Features', fontsize=18, y=1)
# Compute the correlation matrix
corr = df_N.corr()
# Generate a mask for the upper triangle
mask = np.zeros_like(corr, dtype=np.bool)
mask[np.triu_indices_from(mask)] = True
# Set up the matplotlib figure
f, ax = plt.subplots(figsize=(14, 7))
# Generate a custom diverging colormap
cmap = sns.diverging_palette(220, 10, as_cmap=True)
# Draw the heatmap with the mask and correct aspect ratio
sns.heatmap(corr, mask=mask, cmap=cmap, center=0,
square=True, linewidths=.5, cbar_kws={"shrink": .5});
ax.axes.set_title("Correlation Matrix", fontsize=18, y=1.01);
def knn_class(n_iter, n_nbrs, t_size, df_feat, df_target, best_feat_how=None, scaler=None, **best_feat_kwargs):
N = n_iter
train_accuracy_list = []
test_accuracy_list = []
for i in range(N):
X_train, X_test, y_train, y_test = train_test_split(df_feat, df_target, test_size=t_size)
if scaler == 'ss':
scale = StandardScaler()
X_train = scale.fit_transform(X = X_train)
X_test = scale.transform(X = X_test)
elif scaler == 'mm':
scale = MinMaxScaler()
X_train = scale.fit_transform(X_train)
X_test = scale.transform(X_test)
else:
pass
training_accuracy = []
test_accuracy = []
neighbors_settings = range(1, min(n_nbrs, len(X_train)))
for n_neighbors in neighbors_settings:
# build the model
clf = KNeighborsClassifier(n_neighbors=n_neighbors)
clf.fit(X_train, y_train)
# record training set accuracy
training_accuracy.append(clf.score(X_train, y_train))
# record generalization accuracy
test_accuracy.append(clf.score(X_test, y_test))
train_accuracy_list.append(training_accuracy)
test_accuracy_list.append(test_accuracy)
pass
train_mean = np.array(train_accuracy_list).mean(axis=0)
train_stdev = np.array(train_accuracy_list).std(axis=0)/np.sqrt(N)
test_mean = np.array(test_accuracy_list).mean(axis=0)
test_stdev = np.array(test_accuracy_list).std(axis=0)/np.sqrt(N)
neighbors_settings_max = np.argmax(test_mean)+1
test_mean_max = np.round(np.max(test_mean), 4)
if not(best_feat_how):
return ['KNN Classifier', test_mean_max*100, f'k = {neighbors_settings_max}', scaler]
else:
acc_best = 0 if best_feat_how=='indiv' else 100
feat_best = 'placeholder'
for feat in df_feat.columns:
cols = [feat,] if best_feat_how=='indiv' else [i for i in df_feat.columns if i!=feat]
niters = best_feat_kwargs.get('n_iter_best_feat', n_iter)
nnbrs = best_feat_kwargs.get('n_nbrs_best_feat', n_nbrs)
result = knn_class(niters, nnbrs,
0.25, df_feat[cols], df_target)
if (result[1]>acc_best and best_feat_how=='indiv') or (result[1]<acc_best and best_feat_how=='without'):
acc_best = result[1]
feat_best = feat
return ['KNN Classifier', test_mean_max*100, f'k = {neighbors_settings_max}', scaler, feat_best]knn_class(10, 30, 0.25, df_N_features, df_N_target, best_feat_how=None)['KNN Classifier', 75.44, 'k = 11', None]
def pcc(df_target):
target_count = np.unique(df_target, return_counts=True)[1]
pcc_125 = np.round(1.25*np.sum((target_count/target_count.sum())**2), 4)
return pcc_125pcc(df_R_target)0.6618
def logistic(df_feat, df_target, n_iter, t_size, reg, scaler, top_feats=1):
C = [1e-8, 1e-4, 1e-3, 1e-2, 0.1, 0.2,0.4, 0.75, 1, 1.5, 3, 5, 10, 15, 20, 100, 300, 1000, 5000]
score_train = []
score_test = []
random_states = random.sample(range(100), n_iter)
best_feats = {i:np.zeros(len(df_feat.columns)) for i in C}
best_inter = {i:0 for i in C}
for seed in random_states:
training_accuracy = []
test_accuracy = []
X_train, X_test, y_train, y_test = train_test_split(df_feat, df_target, test_size=t_size, random_state=seed)
if scaler == 'ss':
scale = StandardScaler()
X_train = scale.fit_transform(X = X_train)
X_test = scale.transform(X = X_test)
elif scaler == 'mm':
scale = MinMaxScaler()
X_train = scale.fit_transform(X_train)
X_test = scale.transform(X_test)
else:
pass
for alpha_run in C:
lr = LogisticRegression(C=alpha_run, penalty=reg).fit(X_train, y_train)
training_accuracy.append(lr.score(X_train, y_train))
test_accuracy.append(lr.score(X_test, y_test))
if scaler:
coefs = np.mean(lr.coef_, axis=0)
inter = np.mean(lr.intercept_, axis=0)
best_feats[alpha_run] += coefs / n_iter
best_inter[alpha_run] += inter / n_iter
# best_feats[alpha_run].append(df_feat.columns[np.argmax(np.abs(coefs))])
score_train.append(training_accuracy)
score_test.append(test_accuracy)
score = np.mean(score_test, axis=0)
feat_coeffs = best_feats[C[np.argmax(score)]]
intercept = best_inter[C[np.argmax(score)]]
if scaler:
return [f'Logistic {reg}', np.round(np.amax(score)*100, 4), f'C = {C[np.argmax(score)]}', scaler,
list(zip(df_feat.columns[np.argsort(np.abs(feat_coeffs))[-top_feats:][::-1]],
feat_coeffs[np.argsort(np.abs(feat_coeffs))[-top_feats:][::-1]])), (feat_coeffs, intercept)]
else:
return [f'Logistic {reg}', np.round(np.amax(score)*100, 4), f'C = {C[np.argmax(score)]}', scaler]
n = logistic(df_N_features, df_N_target, 20, 0.25, 'l1', 'mm', top_feats=10)n['Logistic l1',
92.3476,
'C = 3',
'mm',
[('Amount', -9.098392587704907),
('Rate', -3.4135979440681745),
('Term', 1.8046593255352945),
('Cost of Funds', 1.086797954491032),
('FICO', -0.663274598966211),
('Partner Bin_2', -0.5807589355467511),
('Tier', -0.39555558541910035),
('Partner Bin_1', 0.3437602061753951),
('Competition rate', -0.17793518989959722),
('Partner Bin_3', -0.1215966566339388)],
(array([-0.39555559, -0.6632746 , 1.80465933, -9.09839259, -0.17793519,
-3.41359794, 1.08679795, 0.34376021, -0.58075894, -0.12159666]),
-0.04370093455394698)]
def y_func(coeff, inter, X):
return np.dot(X,coeff.T) + inter coeff = n[5][0]
inter = n[5][1]X = df_N_features.to_numpy()
X = MinMaxScaler().fit_transform(X)
Xarray([[0. , 0.57751938, 0. , ..., 1. , 0. ,
0. ],
[0. , 0.6124031 , 0.66666667, ..., 1. , 0. ,
0. ],
[0.33333333, 0.50387597, 0.66666667, ..., 0. , 0. ,
1. ],
...,
[0.66666667, 0.63953488, 0.66666667, ..., 1. , 0. ,
0. ],
[0.33333333, 0.44573643, 0.33333333, ..., 0. , 1. ,
0. ],
[0.66666667, 0.76356589, 0.66666667, ..., 0. , 0. ,
1. ]])
y = y_func(coeff, inter, X)
yarray([-2.02880516, -2.70801419, -1.63324091, ..., -3.02602811,
-4.11357686, -2.80025046])
def p(y):
return 1/(1+np.exp(-y))plt.scatter(y, p(y), c=df_N_target.to_numpy(), alpha=0.9)<matplotlib.collections.PathCollection at 0x2182c1fcb38>
c=[df_N_target.to_numpy()==0]
plt.scatter(y[c], p(y)[c])<matplotlib.collections.PathCollection at 0x2182bb430b8>
c=[df_N_target.to_numpy()==1]
plt.scatter(y[c], p(y)[c])<matplotlib.collections.PathCollection at 0x2182d4944e0>
print('$\hat{y} = $')
for i,j in list(zip(df_N_features.columns[np.argsort(np.abs(coeff))[::-1]],
coeff[np.argsort(np.abs(coeff))[::-1]])):
print(f'<br>${j:.3f} x {i}$')
print(f'\t{inter:.3f}')$\hat{y} = $
<br>$-9.135 x Amount$
<br>$-3.426 x Rate$
<br>$1.811 x Term$
<br>$1.096 x Cost of Funds$
<br>$-0.665 x FICO$
<br>$-0.622 x Partner Bin_2$
<br>$-0.388 x Tier$
<br>$0.312 x Partner Bin_1$
<br>$-0.188 x Competition rate$
<br>$-0.151 x Partner Bin_3$
0.037
import seaborn as sns
fig = plt.figure(figsize=(16,9))
c0=[df_N_target.to_numpy()==0]
c1=[df_N_target.to_numpy()==1]
sns.distplot(p(y)[c0], label='Rejected')
sns.distplot(p(y)[c1], label='Availed')
plt.legend()
plt.xlabel('$P(\hat{y})$')
plt.ylabel('Distribution')
plt.title('Logit Probability Distribution of Availed vs Rejected Loans')
#sns.distplot(p(y))Text(0.5, 1.0, 'Logit Probability Distribution of Availed vs Rejected Loans')
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
fig = plt.figure(figsize=(16,9))
ax = plt.axes(projection='3d')
rand = random.sample(range(len(y)),18000)
ax1 = ax.scatter3D(df_N_features['Amount'].to_numpy()[rand], df_N_features['Rate'].to_numpy()[rand], p(y)[rand],
c=df_N_target.to_numpy()[rand], alpha=0.5)
ax.view_init(azim=30)
ax.set_xlabel('Amount (USD)')
ax.set_ylabel('Rate')
ax.set_zlabel('$P(\hat{y})$')
ax.legend(*ax1.legend_elements(), title='Rejected/Availed');
fig = plt.figure(figsize=(16,9))
ratedel = [0, 20, 70]
c0=[df_N_target.to_numpy()==0]
c1=[df_N_target.to_numpy()==1]
for r in ratedel:
x = df_N_features.to_numpy()
x = MinMaxScaler().fit_transform(x)
x[:,df_N_features.columns.values.tolist().index('Rate')] *= (100-r)/100
ypr = y_func(coeff, inter, x)
sns.distplot(p(ypr)[c0], label=str(r)+'% Reduction')
plt.legend()
plt.xlabel('$P(\hat{y})$')
plt.ylabel('Distribution')
plt.title('Shifted Logit Probability Distribution of Rejected Loans')Text(0.5, 1.0, 'Shifted Logit Probability Distribution of Rejected Loans')
u = logistic(df_U_features, df_U_target, 20, 0.25, 'l1', 'mm', top_feats=10)coeff = u[5][0]
inter = u[5][1]
X = df_U_features.to_numpy()
X = MinMaxScaler().fit_transform(X)
y = y_func(coeff, inter, X)print('$\hat{y} = $')
for i,j in list(zip(df_N_features.columns[np.argsort(np.abs(coeff))[::-1]],
coeff[np.argsort(np.abs(coeff))[::-1]])):
print(f'<br>${j:.3f} x {i}$')
print(f'\t{inter:.3f}')$\hat{y} = $
<br>$-13.024 x Amount$
<br>$-9.702 x Rate$
<br>$1.989 x Term$
<br>$1.924 x Partner Bin_1$
<br>$1.749 x Tier$
<br>$1.546 x Partner Bin_3$
<br>$1.362 x Competition rate$
<br>$1.043 x Partner Bin_2$
<br>$-1.010 x FICO$
<br>$-0.193 x Cost of Funds$
1.279
import seaborn as sns
fig = plt.figure(figsize=(16,9))
c0=[df_U_target.to_numpy()==0]
c1=[df_U_target.to_numpy()==1]
sns.distplot(p(y)[c0], label='Rejected')
sns.distplot(p(y)[c1], label='Availed')
plt.legend()
plt.xlabel('$P(\hat{y})$')
plt.ylabel('Distribution')
plt.title('Logit Probability Distribution of Availed vs Rejected Loans')
#sns.distplot(p(y))Text(0.5, 1.0, 'Logit Probability Distribution of Availed vs Rejected Loans')
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
fig = plt.figure(figsize=(16,9))
ax = plt.axes(projection='3d')
rand = random.sample(range(len(y)),18000)
ax1 = ax.scatter3D(df_U_features['Amount'].to_numpy()[rand], df_U_features['Rate'].to_numpy()[rand], p(y)[rand],
c=df_U_target.to_numpy()[rand], alpha=0.5)
ax.view_init(azim=30)
ax.set_xlabel('Amount (USD)')
ax.set_ylabel('Rate')
ax.set_zlabel('$P(\hat{y})$')
ax.legend(*ax1.legend_elements(), title='Rejected/Availed');
fig = plt.figure(figsize=(16,9))
ratedel = [0, 20, 50]
c0=[df_U_target.to_numpy()==0]
c1=[df_U_target.to_numpy()==1]
for r in ratedel:
x = df_U_features.to_numpy()
x = MinMaxScaler().fit_transform(x)
x[:,df_U_features.columns.values.tolist().index('Rate')] *= (100-r)/100
ypr = y_func(coeff, inter, x)
sns.distplot(p(ypr)[c0], label=str(r)+'% Reduction')
plt.legend()
plt.xlabel('$P(\hat{y})$')
plt.ylabel('Distribution')
plt.title('Shifted Logit Probability Distribution of Rejected Loans')Text(0.5, 1.0, 'Shifted Logit Probability Distribution of Rejected Loans')
r = logistic(df_R_features, df_R_target, 20, 0.25, 'l1', 'mm', top_feats=10)coeff = r[5][0]
inter = r[5][1]
X = df_R_features.to_numpy()
X = MinMaxScaler().fit_transform(X)
y = y_func(coeff, inter, X)print('$\hat{y} = $')
for i,j in list(zip(df_R_features.columns[np.argsort(np.abs(coeff))[::-1]],
coeff[np.argsort(np.abs(coeff))[::-1]])):
print(f'<br>${j:.3f} x {i}$')
print(f'<br>{inter:.3f}')$\hat{y} = $
<br>$-5.366 x Rate$
<br>$4.230 x Previous Rate$
<br>$-2.067 x Amount$
<br>$-1.878 x FICO$
<br>$-1.078 x Partner Bin_2$
<br>$1.065 x Competition rate$
<br>$-0.803 x Tier$
<br>$0.646 x Partner Bin_1$
<br>$0.590 x Term$
<br>$0.455 x Cost of Funds$
<br>$0.432 x Partner Bin_3$
<br>0.082
import seaborn as sns
fig = plt.figure(figsize=(16,9))
c0=[df_R_target.to_numpy()==0]
c1=[df_R_target.to_numpy()==1]
sns.distplot(p(y)[c0], label='Rejected')
sns.distplot(p(y)[c1], label='Availed')
plt.legend()
plt.xlabel('$P(\hat{y})$')
plt.ylabel('Distribution')
plt.title('Logit Probability Distribution of Availed vs Rejected Loans')
#sns.distplot(p(y))Text(0.5, 1.0, 'Logit Probability Distribution of Availed vs Rejected Loans')
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
fig = plt.figure(figsize=(16,9))
ax = plt.axes(projection='3d')
rand = random.sample(range(len(y)),18000)
ax1 = ax.scatter3D(df_R_features['Amount'].to_numpy()[rand], df_R_features['Rate'].to_numpy()[rand], p(y)[rand],
c=df_R_target.to_numpy()[rand], alpha=0.5)
ax.view_init(azim=30)
ax.set_xlabel('Amount (USD)')
ax.set_ylabel('Rate')
ax.set_zlabel('$P(\hat{y})$')
ax.legend(*ax1.legend_elements(), title='Rejected/Availed');
fig = plt.figure(figsize=(16,9))
ratedel = [0, 20, 50]
c0=[df_R_target.to_numpy()==0]
c1=[df_R_target.to_numpy()==1]
for r in ratedel:
x = df_R_features.to_numpy()
x = MinMaxScaler().fit_transform(x)
x[:,df_R_features.columns.values.tolist().index('Rate')] *= (100-r)/100
ypr = y_func(coeff, inter, x)
sns.distplot(p(ypr)[c0], label=str(r)+'% Reduction')
plt.legend()
plt.xlabel('$P(\hat{y})$')
plt.ylabel('Distribution')
plt.title('Shifted Logit Probability Distribution of Rejected Loans')Text(0.5, 1.0, 'Shifted Logit Probability Distribution of Rejected Loans')
def svm(df_feat, df_target, n_iter, t_size, reg, scaler, top_feats=1):
C = [1e-8, 1e-4, 1e-3, 1e-2, 0.1, 0.2,0.4, 0.75, 1, 1.5, 3, 5, 10, 15, 20, 100, 300, 1000, 5000]
score_train = []
score_test = []
random_states = random.sample(range(100), n_iter)
best_feats = {i:np.zeros(len(df_feat.columns)) for i in C}
for seed in random_states:
training_accuracy = []
test_accuracy = []
X_train, X_test, y_train, y_test = train_test_split(df_feat, df_target, test_size=t_size, random_state=seed)
if scaler == 'ss':
scale = StandardScaler()
X_train = scale.fit_transform(X = X_train)
X_test = scale.transform(X = X_test)
elif scaler == 'mm':
scale = MinMaxScaler()
X_train = scale.fit_transform(X_train)
X_test = scale.transform(X_test)
else:
pass
for alpha_run in C:
if reg == 'l1':
svc = LinearSVC(C=alpha_run, penalty=reg, loss='squared_hinge', dual=False).fit(X_train, y_train)
if reg == 'l2':
svc = LinearSVC(C=alpha_run, penalty=reg).fit(X_train, y_train)
training_accuracy.append(svc.score(X_train, y_train))
test_accuracy.append(svc.score(X_test, y_test))
if scaler:
coefs = np.mean(svc.coef_, axis=0)
best_feats[alpha_run] += coefs / len(random_states)
score_train.append(training_accuracy)
score_test.append(test_accuracy)
score = np.mean(score_test, axis=0)
feat_coeffs = best_feats[C[np.argmax(score)]]
if scaler:
return [f'SVM {reg}', np.round(np.amax(score)*100, 4), f'C = {C[np.argmax(score)]}', scaler,
list(zip(df_feat.columns[np.argsort(np.abs(feat_coeffs))[-top_feats:][::-1]],
feat_coeffs[np.argsort(np.abs(feat_coeffs))[-top_feats:][::-1]]))]
else:
return [f'SVM {reg}', np.round(np.amax(score)*100, 4), f'C = {C[np.argmax(score)]}', scaler]svm(df_N_features, df_N_target, 20, 0.25, reg='l2', scaler='mm', top_feats=10)['SVM l2',
68.4522,
'C = 3',
'mm',
[('Amount', -3.065494961160274),
('Rate', -1.2714260932964554),
('Term', 0.7058356945328085),
('Partner Bin_1', 0.4310965344476261),
('FICO', -0.3998762425772919),
('Cost of Funds', 0.39218462259885634),
('Partner Bin_3', 0.23468828579738202),
('Tier', -0.21411255364856285),
('Competition rate', -0.04678933472306601),
('Partner Bin_2', -0.026745927908751067)]]
svm(df_N_features, df_N_target, 20, 0.25, reg='l2', scaler='ss', top_feats=10)['SVM l2',
68.4933,
'C = 3',
'ss',
[('Amount', -0.3300914100759237),
('Term', 0.21582348225869322),
('Rate', -0.18953171603578053),
('Cost of Funds', 0.09723813266559847),
('Partner Bin_2', -0.09400410081091995),
('Partner Bin_1', 0.07822553490922914),
('FICO', -0.07334792873898931),
('Tier', -0.0711939628256175),
('Partner Bin_3', -0.01971340167861528),
('Competition rate', -0.010391847535466682)]]
svm(df_N_features, df_N_target, 20, 0.25, reg='l2', scaler=None)['SVM l2', 92.258, 'C = 0.1', None]
svm(df_N_features, df_N_target, 20, 0.25, reg='l1', scaler='mm')['SVM l1', 92.3329, 'C = 1e-08', 'mm', 'FICO']
svm(df_N_features, df_N_target, 20, 0.25, reg='l1', scaler='ss')['SVM l1', 92.3202, 'C = 1e-08', 'ss', 'FICO']
df_U_target.shape(41816,)