def fit(X, y, theta, alpha, num_iters, auto_stop=False):
"""
fit a dataset with only one parameter (using linear_regression)
return (theta, cost_history)
"""
m = X.shape[0]
tmp = [0, 0]
J_history = []
for iter_ in range(num_iters):
tmp[0] = theta[0] - alpha * 1 / m * sum((predict(X, theta) - y) * 1)
tmp[1] = theta[1] - alpha * 1 / m * sum((predict(X, theta) - y) * X)
theta[0] = tmp[0]
theta[1] = tmp[1]
J_history.append(cost(X, y, theta))
if len(J_history) >= 2:
# test is thecost increase
if J_history[-1] - const.INCREASE_THRESHOLD > J_history[
0]: # the cost increase
logerr(
'wrong value of alpha (%f), the cost inscrease -> stop fit'
% (alpha))
return theta, J_history
# test is the fit is done
if auto_stop:
if J_history[-2] <= J_history[-1] + (const.STOP_THRESHOLD *
alpha):
loginfo('auto stopped at %d iterations' % (iter_))
return theta, J_history
return theta, J_history
def export_theta(filename, theta):
try:
with open(filename, 'w') as f:
json.dump(theta, f)
except (os.NotADirectoryError, os.FileNotFoundError):
logerr('unable to write theta in', filename)
loginfo('export theta in %s' % (filename))
def start_predict(all_args):
theta = get_theta_in_args(all_args)
if theta is None:
exit(1)
loginfo('using theta ->', theta)
if all_args['data_predict']['value'] is None:
try:
all_args['data_predict']['value'] = [int(input('car km (int): '))]
except ValueError:
logerr('km should be an int')
exit(1)
for i in all_args['data_predict']['value']:
print('for %8dkm -> estimated price: %d' % (i, predict(i, theta)))
def setarg(all_args, arg):
if '=' in arg and len(arg.split('=')) == 2:
arg_name, arg_value = arg.split('=')
for key, arg_dict in all_args.items():
if arg_name in arg_dict['argnames']:
if convert_from_str(arg_value, arg_dict['type'])[0]:
arg_dict['value'] = convert_from_str(
arg_value, arg_dict['type'])[1]
return True
logerr('in arg %s -> unable to convert %s to %s' %
(arg, arg_value, str(arg_dict['type'])))
return False
logerr('invalid argument ->', arg)
return False
def import_theta(filename):
if not os.path.isfile(filename):
logerr('cannot import theta: %s is not a file' % (filename))
return None
with open(filename, 'r') as f:
try:
theta_data = json.load(f)
except ValueError:
logerr('cannot import theta')
return None
if not check_theta(theta_data):
return None
loginfo('import theta from %s' % (filename))
return theta_data
def check_theta(theta):
"""
check theta values
"""
if type(theta) is not list:
logerr('theta is not a list %s' % (theta))
return False
if len(theta) is not 2:
logerr('invalid theta size (%d: excpected 2) -> %s' %
(len(theta), theta))
return False
try:
theta[0] = float(theta[0])
theta[1] = float(theta[1])
except (ValueError, TypeError):
logerr('cannot convert theta values to float %s' % (theta))
return False
return True
def import_data(all_args):
try:
data = pd.read_csv(all_args['data_filename']['value'])
except (pd.errors.ParserError, pd.errors.EmptyDataError):
logerr('unable to read the csv:', all_args['data_filename']['value'])
return None, None, None
try:
X = np.array(data[all_args['data_km']['value']])
except KeyError:
logerr(
'invalid column %s in the csv: %s' %
(all_args['data_km']['value'], all_args['data_filename']['value']))
return None, None, None
try:
y = np.array(data[all_args['data_price']['value']])
except KeyError:
logerr('invalid column %s in the csv: %s' %
(all_args['data_price']['value'],
all_args['data_filename']['value']))
return None, None, None
return data, X, y
logerr('km should be an int')
exit(1)
for i in all_args['data_predict']['value']:
print('for %8dkm -> estimated price: %d' % (i, predict(i, theta)))
if __name__ == '__main__':
for arg in sys.argv[1:]:
if arg in ['--usage']:
args.print_usage(all_args, sys.argv[0])
exit(0)
elif not args.setarg(all_args, arg):
exit(1)
if all_args['theta']['value'] is not None and not check_theta(
all_args['theta']['value']):
exit(1)
if all_args['data_predict']['value'] is not None:
if type(all_args['data_predict']['value']) is int:
all_args['data_predict']['value'] = [
all_args['data_predict']['value']
]
for data in all_args['data_predict']['value']:
if type(data) is not int:
logerr('we can predict only with int data')
exit(1)
start_predict(all_args)