def regularized_logistic_regression_bounded(log_func, learning_rate, bound,
stop_threshold, pts, l):
curr_cost = regularized_cost(log_func, pts,
squared_coefficient_regularization, l)
print("Initial cost is: {}".format(curr_cost))
increasing_count = 0
consecutive_small_count = 0
itcount = 0
while True:
itcount += 1
gradvals = regularized_gradient_descent_value(
log_func, pts, squared_coefficient_regularization_gradient, l)
nextparams = next_params(log_func, learning_rate, gradvals)
log_func = logistic_function(nextparams)
last_cost, curr_cost = curr_cost, regularized_cost(
log_func, pts, squared_coefficient_regularization, l)
diff = last_cost - curr_cost
# print("Cost after iteration {} is : {} with difference {}".format(itcount, curr_cost, diff))
# print("Params at iteration {} are: {}".format(itcount, log_func["params"]))
if 0 < diff < bound:
consecutive_small_count += 1
elif diff < 0:
increasing_count += 1
if consecutive_small_count > stop_threshold:
break
elif increasing_count > stop_threshold:
raise OverflowError
return log_func
def ols_regression_bounded(lin_func, learning_rate, bound, stop_threshold, pts):
curr_cost = non_regularized_cost(lin_func, pts)
print("Initial cost is: {}".format(curr_cost))
increasing_count = 0
consecutive_small_count = 0
itcount = 0
while True:
itcount += 1
gradvals = non_regularized_gradient_descent_value(lin_func, pts)
nextparams = next_params(lin_func, learning_rate, gradvals)
lin_func = linear_function(nextparams)
last_cost, curr_cost = curr_cost, non_regularized_cost(lin_func, pts)
diff = last_cost - curr_cost
#print("Cost after iteration {} is : {} with difference {}".format(itcount, curr_cost, diff))
if 0 < diff < bound:
consecutive_small_count += 1
elif diff < 0:
increasing_count += 1
if consecutive_small_count > stop_threshold:
break
elif increasing_count > stop_threshold:
raise OverflowError
return lin_func
def logistic_regression(log_func, iterations, learning_rate, pts):
for itcount in range(iterations):
gradvals = non_regularized_gradient_descent_value(log_func, pts)
nextparams = next_params(log_func, learning_rate, gradvals)
log_func = logistic_function(tuple(nextparams))
curr_cost = non_regularized_cost(log_func, pts)
print("Cost after iteration {} is : {}".format(itcount, curr_cost))
return log_func
def ridge_regression(lin_func, iterations, learning_rate, pts, l):
for itcount in range(iterations):
gradvals = regularized_gradient_descent_value(lin_func, pts, squared_coefficient_regularization_gradient, l)
nextparams = next_params(lin_func, learning_rate, gradvals)
lin_func = linear_function(tuple(nextparams))
curr_cost = regularized_cost(lin_func, pts, squared_coefficient_regularization, l)
#print("Cost after iteration {} is : {}".format(itcount, curr_cost))
return lin_func