def interval_regression(df_feature_full, train_begin, test_day):
# make label as int
df_feature_full['result'] = df_feature_full['result'].astype(int)
storeid_list = df_feature_full['storeid'].unique()
goodsn_list = df_feature_full['goodscode'].unique()
df_results_out = pd.DataFrame()
df_corr = pd.DataFrame()
for store_id in storeid_list:
for goodsn in goodsn_list:
# filter data
df_goods_full = df_feature_full[
(df_feature_full['goodscode'] == goodsn)
& (df_feature_full['storeid'] == store_id)
& (df_feature_full['rundate'] >= train_begin)
& (df_feature_full['rundate'] <= test_day)
].drop('goodscode', 1).drop('storeid', 1)
if len(df_goods_full) == 0:
continue
# binning
df_goods_full = f_binning.binner_engine(df_goods_full)
# get feature and label
df_feature = df_goods_full.drop('result', 1)
df_label = df_goods_full[['result', 'rundate']]
# get best parameters
# paras = get_paras(pred_dts[0], df_feature, df_label)
paras = {}
# single process predict
res_stock_list = []
res_sale_list = []
res_sale_list.append(sale_job(paras, test_day, df_feature, df_label))
res_stock_list.append(stock_job(test_day, df_feature, df_label))
# # generate plt draw data
# df_draw = df_goods_full
# df_draw = df_draw[df_draw['rundate'] >
# (datetime.strptime(test_begin, "%Y-%m-%d") - timedelta(days=20)).strftime('%Y-%m-%d')]
# # exporter
# if not print_evaluation(goodsn, res_list):
# continue
# if not plt_draw(store_id, goodsn, df_draw, res_list):
# continue
succ, result = gen_predicts(store_id, goodsn, res_sale_list, res_stock_list)
if not succ:
continue
#result
df_results_out = pd.concat([df_results_out, result], ignore_index=True)
#feature corr output
df_corr = corr_evaluation(df_corr, store_id, goodsn, df_goods_full)
write_corr(df_corr)
return df_results_out
def day_regression(store_id, goodsn_list, df_feature_full, date_max, gap=30):
# make label as int
df_feature_full['result'] = df_feature_full['result'].astype(int)
df_results = pd.DataFrame()
df_corr = pd.DataFrame()
for goodsn in goodsn_list:
df_goods_full = df_feature_full[
(df_feature_full['goodscode'] == goodsn)
& (df_feature_full['storeid'] == store_id)].drop('goodscode',
1).drop(
'storeid', 1)
if len(df_goods_full) == 0:
continue
# binning
df_goods_full = f_binning.binner_engine(df_goods_full)
# feature and label
df_feature = df_goods_full.drop('result', 1)
df_label = df_goods_full[['result', 'rundate']]
# get date range
pred_dts = [
x.strftime('%Y-%m-%d')
for x in pd.date_range(end=date_max, periods=gap).tolist()
]
# get best parameters
dt_min = pred_dts[0]
paras = get_paras(dt_min, df_feature, df_label)
# paras = {}
# single process predict
res_list = []
for pred_dt in pred_dts:
res_list.append(proba_job(paras, pred_dt, df_feature, df_label))
# generate plt draw data
df_draw = df_feature
df_draw['rundate'] = df_draw['rundate'] + timedelta(days=-2)
df_draw = df_draw[df_draw['rundate'] > (
datetime.strptime(date_max, "%Y-%m-%d") -
timedelta(days=gap + 20)).strftime('%Y-%m-%d')]
#exporter
if not print_evaluation(goodsn, res_list):
continue
if not plt_draw(store_id, goodsn, df_draw, res_list):
continue
succ, result = gen_predicts(goodsn, res_list)
if not succ:
continue
df_results = pd.concat([df_results, result], ignore_index=True)
#feature corr output
df_corr = corr_evaluation(df_corr, store_id, goodsn, df_goods_full)
write_corr(df_corr)
return df_results
def day_regression(store_id, goodsn_list, df_feature_full, date_max, gap=30):
# make label as int
df_feature_full['result'] = df_feature_full['result'].astype(int)
df_results = pd.DataFrame()
for goodsn in goodsn_list:
df_goods_full = df_feature_full[
(df_feature_full['goodscode'] == goodsn)
& (df_feature_full['storeid'] == store_id)].drop('goodscode',
1).drop(
'storeid', 1)
if len(df_goods_full) == 0:
continue
# binning
df_goods_full = f_binning.binner_engine(df_goods_full)
# get date range
pred_dts = [
x.strftime('%Y-%m-%d')
for x in pd.date_range(end=date_max, periods=gap).tolist()
]
# single process predict
res_list = []
for pred_dt in pred_dts:
day_ago = get_n_day_ago(pred_dt, 7)
df_goods_full = df_goods_full[df_goods_full['rundate'] > day_ago]
# feature and label
df_feature = df_goods_full.drop('result', 1)
df_label = df_goods_full[['result', 'rundate']]
res_list.append(proba_job(pred_dt, df_feature, df_label))
#exporter
if not print_evaluation(goodsn, res_list):
continue
succ, result = gen_predicts(goodsn, res_list)
if not succ:
continue
df_results = pd.concat([df_results, result], ignore_index=True)
return df_results
def day_regression(store_id, goodsn_list, df_feature_full, gap=30):
df_feature_full['result'] = df_feature_full['result'].astype(int)
df_results = []
for goodsn in goodsn_list:
df_data = df_feature_full[(df_feature_full['goodscode']==goodsn) & (df_feature_full['storeid']==store_id)]\
.drop('goodscode', 1).drop('storeid', 1)
df_simple = df_data.drop('result', 1)
df_label = df_data[['result', 'rundate']]
# get date range
rundate_max = df_simple['rundate'].max()
if rundate_max == 'nan' or str(rundate_max) == 'NaT':
continue
df_simple = f_binning.binner_engine(df_simple)
df_max = rundate_max.strftime('%Y-%m-%d')
pred_dts = [
x.strftime('%Y-%m-%d')
for x in pd.date_range(end=df_max, periods=gap).tolist()
]
# single process
res_list = []
for pred_dt in pred_dts:
res_list.append(proba_job(pred_dt, df_simple, df_label))
# Arrangement data
df_draw = df_simple
df_draw['rundate'] = df_draw['rundate'] + timedelta(days=-2)
df_draw = df_draw[df_draw['rundate'] > (
rundate_max - timedelta(days=gap)).strftime('%Y-%m-%d')]
y_tests = []
y_preds = []
dts = []
for res in res_list:
if not res:
continue
y_tests.append(res['y_test'])
y_preds.append(res['y_pred'])
dts.append(res['rundate'])
if len(y_preds) == 0:
continue
y_preds = np.array(y_preds)
y_tests = np.array(y_tests)
print('goodsn: ' + str(goodsn))
print('\tmean(stock): ' + str(df_draw['endqty'].mean()))
print('\tmean(sales): ' + str(y_tests.mean()))
print('\tmean(predict_stock): ' + str(y_preds.mean()))
print('\tcount(stock<0): ' + str(len(df_draw[df_draw['endqty'] < 0])))
print('\tpredict_stock<sales: ' + str(np.sum(y_preds < y_tests)))
print('\tcount(sales): ' + str(len(y_tests)))
# draw sales chart
plt.figure(1, figsize=(20, 7))
plt.plot(df_draw['rundate'],
df_draw['saleqty'],
color='#6aa84f',
marker='o')
# plt.plot(df_draw['rundate'], df_draw['stockqty'], color='black', marker='o')
plt.plot([datetime.strptime(d, '%Y-%m-%d').date() for d in dts],
y_preds,
color='red',
marker='o')
plt.savefig('../sources/' + str(store_id) + '_' + str(goodsn) + '.png')
plt.close('all')
df_result = pd.DataFrame(dts, columns=['rundate'])
df_result['pred'] = y_preds
df_result['sales'] = y_tests
df_result['goodsn'] = goodsn
df_results.append(df_result)
df_results_detail = pd.concat(df_results, ignore_index=True)
return df_results_detail