def exp1(): # {{{
# data before 2013 to drop or not
d = RowDataHandler()
for filename, inpu_method in data_common.items():
d.add(pd.read_csv(f'{path}/{filename}.csv'), inpu_method)
for filename, inpu_method in data_wu.items():
d.add(pd.read_csv(f'{path}/wu/{filename}.csv'), inpu_method)
start, end = d.get_start_end_tick()
merged_data = d.get_merged_data(start, end)
data, ans, date = merged_data.iloc[:, 1:], merged_data['wu_day_price'],\
merged_data['date']
data_retain, _ = preprocess(data, ans, date, 1, 1, 7, 1)
merged_data = d.get_merged_data('2014-01-01', end)
data_drop, _ = preprocess(merged_data, 1, 1, 7, 1)
retain = []
drop = []
for i in range(10):
res_retain = train_and_eval_model('test',
'large',
**data_retain,
drop_model=True)
res_drop = train_and_eval_model('test',
'large',
**data_drop,
drop_model=True)
retain.append(res_retain)
drop.append(res_drop)
print('retain data before 2013:', retain)
print('drop data before 2013:', drop)
def exp6(): # {{{
input_data, ans, date = load_wu([
'SAR', 'yb_price', 'pa', 'humidity_low', 'wind_max_dir',
'wu_day_price', 'wu_day_amount'
])
for input_size in range(1, 15):
mses = []
for i in range(5):
processed_data = preprocess(input_data, ans, date, 30, 30,
input_size, 1)
data, _ = split(processed_data,
tr_ratio=0.8,
va_ratio=0.1,
te_ratio=0.1)
npad = ((0, 0), (0, 3), (0, 0))
data['train_x'] = np.pad(data['train_x'],
npad,
'constant',
constant_values=0)
data['valid_x'] = np.pad(data['valid_x'],
npad,
'constant',
constant_values=0)
data['test_x'] = np.pad(data['test_x'],
npad,
'constant',
constant_values=0)
mse = train_and_eval_model('test', 'base', **data, drop_model=True)
mses.append(mse)
print(f'inupt_size={input_size}\nmse={np.mean(mse)}')
def exp3(): # {{{
# testing for results of exp2, results of selection
target = argv[2]
if 'wu' == target:
input_data, ans, date = load_wu()
elif 'chi' == target:
input_data, ans, date = load_chi()
else:
raise Exception(f'Error target: {target}')
with open(f'logs/selec_res_{target}', 'r') as f:
cols_set = [
line.strip()[11:-1].replace('\'', '').split(' ')
for line in f.readlines()
]
for use_cols in cols_set:
input_data = input_data[use_cols]
processed_data = preprocess(input_data, ans, date, 1, 1, 7, 1)
data, _ = split(processed_data,
tr_ratio=0.8,
va_ratio=0.1,
te_ratio=0.1)
res_base = []
res_large = []
for i in range(5):
res1 = train_and_eval_model('test',
'base',
**data,
drop_model=True)
res2 = train_and_eval_model('test',
'large',
**data,
drop_model=True)
res_base.append(res1)
res_large.append(res2)
with open(f'logs/exp3_eval_selection_{target}', 'a+') as f:
f.write(f'1.\n')
f.write(f'Use columns: {use_cols}\n')
f.write(f'base model: {res_base}\n')
f.write(f'base mean: {np.mean(res_base)}\n')
f.write(f'large model: {res_large}\n')
f.write(f'large mean: {np.mean(res_large)}\n\n')
def test():
input_data, ans, date = load_wu([
'AED', 'pf_price', 'temp', 'humidity', 'wind_speed', 'wu_ex_weight',
'wu_ex_price', 'wu_month_price'
])
processed_data = preprocess(input_data, ans, date, 1, 1, 7, 1)
data, _ = split(processed_data, tr_ratio=0.8, va_ratio=0.1, te_ratio=0.1)
res = train_and_eval_model('test', 'base', **data)
print(res)
def exp2(): # {{{
# which features to use
if 'wu' == argv[2]:
input_data, ans, date = load_wu()
elif 'chi' == argv[2]:
input_data, ans, date = load_chi()
else:
exit()
columns = input_data.columns.to_numpy()
columns_history = []
mse_history = []
ori_columns = columns.copy()
for _ in range(len(columns) - 2):
res_list = []
for i in range(len(columns)):
mask = np.ones(len(columns), dtype=bool)
mask[i] = False
data_mask = columns[mask]
processed_data = preprocess(input_data[data_mask], ans, date, 1, 1,
7, 1)
data, _ = split(processed_data,
tr_ratio=0.8,
va_ratio=0.1,
te_ratio=0.1)
res = train_and_eval_model('test', 'base', **data, drop_model=True)
res_list.append(res)
idx = np.argmin(res_list)
columns = np.delete(columns, idx)
columns_history.append(columns)
mse_history.append(res_list)
print('The origin columns: ')
print(ori_columns)
print('The final columns: ')
print(columns)
for i in range(len(columns_history)):
print('res: ', mse_history[i])
print('columns: ', columns_history[i])
print('')
def main(): # {{{
path = './data/use'
# value is None for daily data
data_common = {
'currency': None,
'powder_feed': ['same', 'same'],
'yellow_bean': ['same'],
'weather': None
}
data_wu = {
'wu_export': ['divide', 'same'],
'wu_price_perDate': None,
# 'wu_price_perMonth': ['same', 'divide'],
}
data_chi = {
'chi_export': ['divide', 'same'],
'chi_price_perDate': None,
'chi_price_perMonth': ['same', 'divede'],
'chi_small_fish': ['divide', 'same']
}
d = RowDataHandler()
for filename, inpu_method in data_common.items():
d.add(pd.read_csv(f'{path}/{filename}.csv'), inpu_method)
for filename, inpu_method in data_wu.items():
d.add(pd.read_csv(f'{path}/wu/{filename}.csv'), inpu_method)
# ['date', 'USD', 'CAD', 'SAR', 'AED', 'pf_weight', 'pf_price', 'yb_price', 'temp', 'temp_high', 'temp_low', 'point_temp', 'wu_ex_weight', 'wu_ex_price', 'wu_day_price', 'wu_day_amount']
merged_data = d.get_merged_data(*d.get_start_end_tick())[[
'date', 'wu_day_price'
]]
data, time = preprocess(merged_data, 1, 1, 7, 1)
linear_regression(data['test_x'], data['test_y'])
merged_data = d.get_merged_data(*d.get_start_end_tick())
merged_data = merged_data[[
'date', 'temp', 'temp_high', 'temp_low', 'wu_ex_weight', 'wu_ex_price',
'wu_day_price', 'wu_day_amount'
]]
merged_data = merged_data.dropna().reset_index(drop=True)
data, time = preprocess(merged_data, 1, 1, 7, 1)
train_and_eval_model('weather_7dl', 'large', **data)
merged_data = d.get_merged_data(*d.get_start_end_tick())
merged_data = merged_data[[
'date', 'temp', 'temp_high', 'temp_low', 'wu_ex_weight', 'wu_ex_price',
'wu_day_price', 'wu_day_amount'
]]
merged_data = merged_data.dropna().reset_index(drop=True)
data, time = preprocess(merged_data, 1, 1, 14, 1)
train_and_eval_model('weather_14dl', 'large', **data)
merged_data = d.get_merged_data(*d.get_start_end_tick())
merged_data = merged_data[[
'date', 'temp', 'temp_high', 'temp_low', 'wu_ex_weight', 'wu_ex_price',
'wu_day_price', 'wu_day_amount'
]]
merged_data = merged_data.dropna().reset_index(drop=True)
data, time = preprocess(merged_data, 7, 7, 4, 1)
train_and_eval_model('weather_4wl', 'large', **data)
merged_data = d.get_merged_data(*d.get_start_end_tick())
merged_data = merged_data[[
'date', 'temp', 'temp_high', 'temp_low', 'wu_ex_weight', 'wu_ex_price',
'wu_day_price', 'wu_day_amount'
]]
merged_data = merged_data.dropna().reset_index(drop=True)
data, time = preprocess(merged_data, 30, 30, 4, 1)
train_and_eval_model('weather_4ml', 'large', **data)
processed_data = preprocess(input_data, ans, date, 1, 1, 7, 1)
data, _ = split(processed_data, tr_ratio=0.8, va_ratio=0.1, te_ratio=0.1)
res = train_and_eval_model('test', 'base', **data)
print(res)
if '__main__' == __name__:
if 1 == len(argv):
test()
exit()
if '1' == argv[1]:
exp1()
elif '2' == argv[1]:
exp2()
elif '3' == argv[1]:
exp3()
elif '4' == argv[1]:
exp4()
elif 'wu' == argv[1] or 'chi' == argv[1]:
target = argv[1]
mode = argv[2]
used_model = 'large'
data, _ = load_data(target, mode)
model_name = f'{target}_{mode}_{used_model}'
res = train_and_eval_model(model_name, used_model, **data)
print(f'model {model_name}: ', res)
