def templateROC(c='c002812'):
data = pd.read_csv(DATA + 'roc_curve_' + c + '_small.csv')
# data_train, data_test = train_test_split(data, test_size=1.0/20, random_state=101)
# data_test = data_test.sort_values(by=['tpr', 'fpt', 'threshold'])
# data_test.to_csv(DATA + 'roc_curve_' + c + '_small.csv', index=False)
source = ColumnDataSource(data=dict(tpr=data['tpr'],
fpr=data['fpt'],
thre=data['threshold']
))
TOOLS = "pan,wheel_zoom,reset,hover,save"
p = Figure(tools=TOOLS, height=300, width=300, toolbar_location="above")
p.line('fpr', 'tpr', source=source, line_width=5)
p.line([0, 1], [0, 1], line_dash='dashed', line_alpha=0.6)
p.yaxis.axis_label = 'True Positive Rate'
p.xaxis.axis_label = 'False Positive Rate'
# p.background_fill_color = LIGHT_GREEN
# p.border_fill_color = LIGHT_GREEN
hover = p.select_one(HoverTool)
hover.point_policy = "follow_mouse"
hover.tooltips = [
("FPR", "@fpr{1.11}"),
("TPR", "@tpr{1.11}"),
("THRESHOLD", "@thre{1.11}")
]
#
script, div = components(p)
return script, div
p2 = Figure(x_range=(0, 1000),
y_range=(0, 10),
plot_width=600,
plot_height=150,
tools='hover',
title='hover over color')
color_range1 = p2.rect(x='x',
y='y',
width=1,
height=10,
color='crcolor',
source=crsource)
# set up hover tool to show color hex code and sample swatch
p2.select_one(HoverTool).tooltips = [('color',
'$color[hex, rgb, swatch]:crcolor'),
('RGB levels', '@RGBs')]
# theme everything for a cleaner look
curdoc().theme = Theme(json=yaml.load("""
attrs:
Plot:
toolbar_location: null
Grid:
grid_line_color: null
Axis:
axis_line_color: null
major_label_text_color: null
major_tick_line_color: null
minor_tick_line_color: null
cry = [ 5 for i in range(len(crx)) ]
crcolor, crRGBs = generate_color_range(1000,brightness) # produce spectrum
# make data source object to allow information to be displayed by hover tool
crsource = ColumnDataSource(data=dict(x=crx, y=cry, crcolor=crcolor, RGBs=crRGBs))
# create second plot
p2 = Figure(x_range=(0,1000), y_range=(0,10),
plot_width=600, plot_height=150,
tools='hover', title='hover over color')
color_range1 = p2.rect(x='x', y='y', width=1, height=10,
color='crcolor', source=crsource)
# set up hover tool to show color hex code and sample swatch
p2.select_one(HoverTool).tooltips = [
('color', '$color[hex, rgb, swatch]:crcolor'),
('RGB levels', '@RGBs')
]
# theme everything for a cleaner look
curdoc().theme = Theme(json=yaml.load("""
attrs:
Plot:
toolbar_location: null
Grid:
grid_line_color: null
Axis:
axis_line_color: null
major_label_text_color: null
major_tick_line_color: null
def templateRateCorrelation(state):
DEFAULT_X = ['Amount Requested', 'Annual Income', 'Debt To Income Ratio']
dati = pd.read_csv(DATA + 'accepted_less_col_small_' + state + '.csv', header=0)
amnt = dati['amnt']
income = dati['income']
dti = dati['dti']
rate = dati['rate'] / 100
source = ColumnDataSource(
data={'x': amnt,
'y': rate,
'Amount Requested': amnt,
'Annual Income': income,
'Debt To Income Ratio': dti,
'Rate_per_100': rate * 100}
)
codex = """
var data = source.get('data');
data['x'] = data[cb_obj.get('value')];//
// var r = data[cb_obj.get('value')];
// var {var} = data[cb_obj.get('value')];
// //window.alert( "{var} " + cb_obj.get('value') + {var} );
// for (i = 0; i < r.length; i++) {{
// {var}[i] = r[i] ;
// data['{var}'][i] = r[i];
// }}
source.trigger('change');
"""
callbackx = CustomJS(args=dict(source=source), code=codex)
TOOLS = "pan,wheel_zoom,reset,hover,save"
plot = Figure(title=None, height=400, width=600, tools=TOOLS)
# Make a line and connect to data source
plot.circle(x="x", y="y", line_color="#0062cc", line_width=6, line_alpha=0.6, source=source)
plot.yaxis.axis_label = 'Loan Rate'
plot.yaxis[0].formatter = NumeralTickFormatter(format="0.0%")
xaxis_select = Select(title="Label X axis:", value="Amount",
options=DEFAULT_X, callback=callbackx)
hover = plot.select_one(HoverTool)
hover.point_policy = "follow_mouse"
hover.tooltips = [
("Rate", "@Rate_per_100{1.11}%"),
("Amount Requested", "@{Amount Requested}{1.11}"),
("Annual Income", "@{Annual Income}{1.11}"),
("Debt To Income Ratio", "@{Debt To Income Ratio}{1.11}%")
]
# Layout widgets next to the plot
controls = VBox(xaxis_select)
layout = HBox(controls, plot, width=800)
# show(layout)
script_corr, div_corr = components(layout)
return script_corr, div_corr
def outputGridSearchLogisticRegression(path=DATA_LOCAL +
'accepted_refused_ds.csv'):
data_file = open(DATA_LOCAL + 'dict_recall.json', 'r')
# print data_file.read()
data = json.load(data_file)
# pprint(data)
params = data['param_C']['data']
acc = data['mean_test_score']
best_param = params[acc.index(max(acc))]
print 'Best param: ' + str(best_param)
# log loss: 0.0281176869797
# f1: 0.0281176869797
# roc_auc: 0.0001
# recall: 0.0001
plt.semilogx(data['param_C']['data'], data['mean_test_score'])
# plt.show()
plt.close()
print 'Data transformation'
# my_data = dataTransformation(path)
my_data = pd.read_csv(DATA_LOCAL + 'accepted_refused_ds_trans.csv',
header=0)
print my_data.head()
y = my_data['loan']
str_a = 'acc'
str_r = 'ref'
y = y.map(lambda x: str(x).replace('1', str_a).replace('0', str_r))
X = my_data.drop('loan', axis=1)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.3,
random_state=101)
# #print sum(y_test)
#print len(y_test) - sum(y_test)
print 'Built and fit the model'
lr = LogisticRegression(class_weight='balanced',
random_state=101,
C=best_param)
lr.fit(X=X_train,
y=y_train.map(
lambda x: int(x.replace(str_a, '1').replace(str_r, '0'))))
y_pred = lr.predict_proba(X=X_test)
coef = lr.coef_
print lr.get_params()
name_coef = my_data.columns.values[:-1]
coef_val = pd.DataFrame(data={'coef': name_coef, 'val': coef[0]})
coef_val.to_csv(DATA_LOCAL + 'LogisticRegressionCoef_c' + str(best_param) +
'.csv',
index=False)
print 'Built ROC curve'
# voglio avere valori = 0 per ACCETTATO
# = 1 per REFUSED
# cos' nella matroce di cinfusione ho TP FN
# FP TN
y_acc = [p[1] for p in y_pred]
y_test = y_test.map(
lambda x: int(x.replace(str_a, '1').replace(str_r, '0')))
fpr, tpr, threshold = roc_curve(y_true=y_test, y_score=y_acc)
# data_roc = pd.DataFrame({'tpr': tpr, 'fpt': fpr, 'threshold': threshold})
# data_roc.to_csv(DATA_LOCAL# + 'roc_curve_c00001.csv', index=False)
l = np.arange(len(tpr))
roc = pd.DataFrame({
'fpr': pd.Series(fpr, index=l),
'tpr': pd.Series(tpr, index=l),
'1-fpr': pd.Series(1 - fpr, index=l),
'tf': pd.Series(tpr - (1 - fpr), index=l),
'thresholds': pd.Series(threshold, index=l)
})
print roc.ix[(roc.tf - 0.0).abs().argsort()[:1]]
# Plot tpr vs 1-fpr
fig, ax = plt.subplots()
plt.plot(roc['tpr'])
plt.plot(roc['1-fpr'], color='red')
plt.xlabel('1-False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('Receiver operating characteristic')
ax.set_xticklabels([])
# plt.show()
roc_auc = auc(fpr, tpr)
plt.figure()
lw = 2
plt.plot(fpr,
tpr,
color='darkorange',
lw=lw,
label='ROC curve (area = %0.8f)' % roc_auc)
plt.plot([0, 1], [0, 1], color='navy', lw=lw, linestyle='--')
plt.xlim([0.0, 1.0])
plt.ylim([0.0, 1.05])
plt.xlabel('False Positive Rate')
plt.ylabel('True Positive Rate')
plt.title('Receiver operating characteristic')
plt.legend(loc="lower right")
# plt.show()
# thre = float(roc.ix[(roc.tf - 0.0).abs().argsort()[:1]]['thresholds'])
thre = 0.40
source = ColumnDataSource(data=dict(tpr=tpr, fpr=fpr, thre=threshold))
TOOLS = "pan,wheel_zoom,reset,hover,save"
p = Figure(tools=TOOLS)
p.line('fpr', 'tpr', source=source, line_width=4)
p.line([0, 1], [0, 1], line_dash='dashed', line_alpha=0.6)
p.yaxis.axis_label = 'True Positive Rate'
p.xaxis.axis_label = 'False Positive Rate'
hover = p.select_one(HoverTool)
hover.point_policy = "follow_mouse"
hover.tooltips = [("FPR", "@fpr{1.11}"), ("TPR", "@tpr{1.11}"),
("THRESHOLD", "@thre{1.11}")]
show(p)
y_pred_05 = [str_a if a > 0.5 else str_r for a in y_acc]
y_pred_thre = [str_a if a > thre else str_r for a in y_acc]
y_test = y_test.map(
lambda x: str(x).replace('1', str_a).replace('0', str_r))
print 'Confusion matrix threshold = ' + str(thre)
print confusion_matrix(y_true=y_test, y_pred=y_pred_thre)
print 'Confusion matrix threshold = 0.5'
print confusion_matrix(y_true=y_test, y_pred=y_pred_05)
y_pred_05 = [1 if a > 0.5 else 0 for a in y_acc]
y_pred_thre = [1 if a > thre else 0 for a in y_acc]
y_test = y_test.map(
lambda x: int(x.replace(str_a, '1').replace(str_r, '0')))
print 'Score f1 (thres = ' + str(thre) + ') = ' + str(
f1_score(y_true=y_test, y_pred=y_pred_thre))
print 'Score f1 (thres = 0.5) = ' + str(
f1_score(y_true=y_test, y_pred=y_pred_05))
print 'Score recall (thres = ' + str(thre) + ') = ' + str(
recall_score(y_true=y_test, y_pred=y_pred_thre))
print 'Score recall (thres = 0.5) = ' + str(
recall_score(y_true=y_test, y_pred=y_pred_05))
print 'Score accuracy (thres = ' + str(thre) + ') = ' + str(
accuracy_score(y_true=y_test, y_pred=y_pred_thre))
print 'Score accuracy (thres = 0.5) = ' + str(
accuracy_score(y_true=y_test, y_pred=y_pred_05))
print 'Score precision (thres = ' + str(thre) + ') = ' + str(
precision_score(y_true=y_test, y_pred=y_pred_thre))
print 'Score precision (thres = 0.5) = ' + str(
precision_score(y_true=y_test, y_pred=y_pred_05))
print 'Score AUC = ' + str(roc_auc)
