def main():
dev = pd.read_pickle("dev.pkl")
test = pd.read_pickle("test.pkl")
features = [
v
for v in dev.columns.tolist()
if v
not in [
"vintage",
"PMI66007Rating",
"BAE201405ProjLoss",
"LossRate",
"PMI66007LossRate",
"M2_DQ15plus_Prin",
"M3_DQ_Prin",
"UTD_DQ_Prin",
"PMI5ApprovalFlag",
]
]
formula = "UTD_DQ15plus_Prin / LoanAmount : LoanAmount ~ " + " + ".join(features)
strb = StrategyRobot(dev, test, formula=formula, aggregate=False)
book = strb.book_creation(nbins=10, nbins_monotone=20, monotone_sig_level_threshold=0.1, yname="UTD 15+ DQ (%)")
book.make_control_plots(
which_variables=features,
ylim=(0, 4),
rounding=1,
canvas_size=(9, 5),
output_dir=os.getcwd() + "/UTD_15plus_DQ_",
)
var_sel = strb.variable_selection(book)
top_variable_book = var_sel.select_k_best(
method="univariate",
drop_correlated=True,
drop_correlated_threshold=0.6,
k_best=10,
force_monotone=True,
rounding=1,
output_dir=os.getcwd() + "/output/UTD_15plus_DQ_",
)
optimize = strb.optimization(top_variable_book, n_init_vars=3, eval_on_penalized=True, patience=True)
optimize.search_rules(
decrease_frac=0.1,
stopping_eff=0.02 # ,
# force_granularity_delta = 0.999
)
optimize.expand_rules(increase_frac=0.1)
optimize.prune_rules(tolerate_n_std=0.0)
optimize.select_rules()
optimize.performance_output(output_dir=os.getcwd() + "/output/UTD_15plus_DQ_")
def main():
dev = pd.read_pickle("data_input.pkl")
#dev = pd.read_pickle("data_input_old.pkl")
#dev = pd.read_pickle("data_input_new.pkl")
test = dev
features = [v for v in dev.columns.tolist() if v not in ['M2_16PlusDQ_Prin', 'state', 'group_label']]
#features = ['FICO', 'PMI6_1']
#features = ['BAC031_NumOpenBankcardTradesBalanceGT0ReptdLast6Mos', 'FICO']
#features = features[300:]
formula = 'M2_16PlusDQ_Prin / loanamount : loanamount | group_label ~ ' + ' + '.join(features)
print(dev.shape)
strb = StrategyRobot(dev, test, formula = formula, aggregate = False)
book = strb.book_creation(nbins = 5,
nbins_monotone = 5,
monotone_sig_level_threshold = 0.4,
yname = '16+ DQ at M2 (%)')
#book.make_control_plots(which_variables = features,
# ylim = (0, 2),
# ylim_double_ratio = (0.5, 3.5),
# rounding = 2,
# canvas_size = (9, 5),
# label_test_group = '11/15 - 02/16',
# label_control_group = '08/15 - 10/15',
# output_dir = os.getcwd() + '/output/16plus_DQ_')
var_sel = strb.variable_selection(book)
top_variable_book = var_sel.select_k_best(method = 'univariate',
drop_correlated = True,
drop_correlated_threshold = 0.6,
k_best = 12,
force_monotone = True,
rounding = 1,
output_dir = os.getcwd() + '/output/16plus_DQ_')
make_top_variable_plot(top_variable_book)