import matplotlib.pyplot as plt import numpy as np from sklearn.linear_model import SGDClassifier from sklearn.metrics import f1_score from MlcLinReg import MlcLinReg from helpers import shuffle_dataset, split_train_test, tic, toc, load_delicious """ This script is used to compare performance of MlcLinReg and sklearn's SGDClassifier. """ feature = 1 # Load shuffle and split data X_train, y_train, X_test, y_test = load_delicious(feature) X_train_s, y_train_s = shuffle_dataset(X_train, y_train) X_train, y_train, X_test, y_test = split_train_test(X_train_s, y_train_s) # N = np.array(range(100, 1000, 10)) # N = np.array([50, 100, 200, 300, 400, 600]) # N = np.array([0.001, 0.005, 0.025, 0.05, 0.1]) N = np.array([5, 32, 64, 100, 128, 200, 256, 350, 512, 700, 1024, 1500, 2048]) scores = list() scores_sgd = list() times = list() times_sgd = list() batch_size = 2048 iterations = 200
"""
This script runs a randomised grid search on all features of delicious dataset
"""
warnings.filterwarnings("ignore")
param_dist = {
"learning_rate": st.uniform(0.001, 0.4),
"iterations": sp_randint(50, 1000),
"batch_size": sp_randint(2, 2000),
"l_one": st.uniform(0.01, 0.5)
}
best_params = np.zeros((501, 4))
# run randomized search
for feature in tqdm.tqdm(range(0, 501)):
X_train, y_train, X_test, y_test = helpers.load_delicious(feature)
clf = MlcLinReg()
n_iter_search = 60
random_search = RandomizedSearchCV(clf,
param_distributions=param_dist,
n_iter=n_iter_search)
start = time()
random_search.fit(X_train.toarray(), y_train.toarray())
conf = helpers.report_params(random_search.cv_results_, n_top=1)
best_params[feature, :] = conf.values()
# 'delicious_best_params' has the following columns :
# learning_rate l_one iterations batch_size
np.savetxt("delicious_best_params.txt", best_params)
import MlcLinReg
import helpers
# for i in range(100, 100):
# helpers.plot_roc_curve(MlcLinReg.MlcLinReg(batch_size=i), savefig=True)
helpers.plot_roc_curve(MlcLinReg.MlcLinReg(batch_size=256),
dataset=helpers.load_delicious(2),
savefig=False)