def test_binary_logistic_regression_gd():
t = np.array([[0.13, -0.12], [-3.07, 3.05]])
lr = SoftmaxRegression(epochs=200, eta=0.005, minibatches=1, random_seed=1)
lr.fit(X_bin, y_bin)
np.testing.assert_almost_equal(lr.w_, t, 2)
assert (y_bin == lr.predict(X_bin)).all()
def test_multi_logistic_regression_gd_acc():
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X, y)
assert (y == lr.predict(X)).all()
def test_multi_logistic_regression_gd_acc():
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X, y)
assert (y == lr.predict(X)).all()
def test_multi_logistic_probas():
lr = SoftmaxRegression(epochs=200, eta=0.005, minibatches=1, random_seed=1)
lr.fit(X, y)
idx = [0, 50, 149] # sample labels: 0, 1, 2
y_pred = lr.predict_proba(X[idx])
exp = np.array([[1.0, 0.0, 0.0], [0.08, 0.60, 0.32], [0.0, 0.00, 0.99]])
np.testing.assert_almost_equal(y_pred, exp, 2)
def test_score_function():
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X, y)
acc = lr.score(X, y)
assert acc == 1.0, acc
def test_score_function():
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X, y)
acc = lr.score(X, y)
assert acc == 1.0, acc
def test_progress_3():
lr = SoftmaxRegression(epochs=1,
eta=0.005,
minibatches=1,
print_progress=3,
random_seed=1)
lr.fit(X_bin, y_bin) # 0, 1 class
def test_progress_2():
lr = SoftmaxRegression(epochs=1,
eta=0.005,
minibatches=1,
print_progress=2,
random_seed=1)
lr.fit(X_bin, y_bin) # 0, 1 class
def test_multi_logistic_regression_gd_weights():
t = np.array([[-0.95, -2.45, 3.4],
[-3.95, 2.34, 1.59]])
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X, y)
np.testing.assert_almost_equal(lr.w_, t, 2)
def test_multi_logistic_regression_gd_weights():
t = np.array([[-0.95, -2.45, 3.4],
[-3.95, 2.34, 1.59]])
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X, y)
np.testing.assert_almost_equal(lr.w_, t, 2)
def test_binary_logistic_regression_sgd():
t = np.array([[-0.68, 0.68], [-3.2, 3.2]])
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=len(y_bin),
random_seed=1)
lr.fit(X_bin, y_bin) # 0, 1 class
np.testing.assert_almost_equal(lr.w_, t, 2)
assert (y_bin == lr.predict(X_bin)).all()
def test_binary_logistic_regression_gd():
t = np.array([[-0.2, 0.2],
[-3.09, 3.09]])
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X_bin, y_bin)
np.testing.assert_almost_equal(lr.w_, t, 2)
assert((y_bin == lr.predict(X_bin)).all())
def test_binary_l2_regularization_gd():
t = np.array([[-0.17, 0.17], [-2.26, 2.26]])
lr = SoftmaxRegression(epochs=200,
eta=0.005,
l2=1.0,
minibatches=1,
random_seed=1)
lr.fit(X_bin, y_bin)
np.testing.assert_almost_equal(lr.w_, t, 2)
assert (y_bin == lr.predict(X_bin)).all()
def test_binary_logistic_regression_sgd():
t = np.array([[0.13, -0.12],
[-3.06, 3.05]])
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=len(y_bin),
random_seed=1)
lr.fit(X_bin, y_bin) # 0, 1 class
np.testing.assert_almost_equal(lr.w_, t, 2)
assert (y_bin == lr.predict(X_bin)).all()
def test_refit_weights():
t = np.array([[0.13, -0.12], [-3.07, 3.05]])
lr = SoftmaxRegression(epochs=100, eta=0.005, minibatches=1, random_seed=1)
lr.fit(X_bin, y_bin)
w1 = lr.w_[0][0]
w2 = lr.w_[0][0]
lr.fit(X_bin, y_bin, init_params=False)
assert w1 != lr.w_[0][0]
assert w2 != lr.w_[1][0]
np.testing.assert_almost_equal(lr.w_, t, 2)
def test_binary_l2_regularization_gd():
t = np.array([[-0.17, 0.17],
[-2.26, 2.26]])
lr = SoftmaxRegression(epochs=200,
eta=0.005,
l2=1.0,
minibatches=1,
random_seed=1)
lr.fit(X_bin, y_bin)
np.testing.assert_almost_equal(lr.w_, t, 2)
assert (y_bin == lr.predict(X_bin)).all()
def test_multi_logistic_probas():
lr = SoftmaxRegression(epochs=200,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X, y)
idx = [0, 50, 149] # sample labels: 0, 1, 2
y_pred = lr.predict_proba(X[idx])
exp = np.array([[0.99, 0.01, 0.00],
[0.01, 0.88, 0.11],
[0.00, 0.02, 0.98]])
np.testing.assert_almost_equal(y_pred, exp, 2)
def test_binary_l2_regularization_gd():
lr = SoftmaxRegression(eta=0.005,
epochs=200,
minibatches=1,
l2_lambda=1.0,
random_seed=1)
lr.fit(X_bin, y_bin)
y_pred = lr.predict(X_bin)
expect_weights = np.array([[-0.316, 0.317], [-2.265, 2.265]])
np.testing.assert_almost_equal(lr.w_, expect_weights, 3)
acc = sum(y_pred == y_bin) / len(y_bin)
assert acc == 1.0
def test_binary_l2_regularization_gd():
lr = SoftmaxRegression(eta=0.005,
epochs=200,
minibatches=1,
l2_lambda=1.0,
random_seed=1)
lr.fit(X_bin, y_bin)
y_pred = lr.predict(X_bin)
expect_weights = np.array([[-0.316, 0.317],
[-2.265, 2.265]])
np.testing.assert_almost_equal(lr.w_, expect_weights, 3)
acc = sum(y_pred == y_bin) / len(y_bin)
assert(acc == 1.0)
def test_refit_weights():
t = np.array([[0.13, -0.12],
[-3.07, 3.05]])
lr = SoftmaxRegression(epochs=100,
eta=0.005,
minibatches=1,
random_seed=1)
lr.fit(X_bin, y_bin)
w1 = lr.w_[0][0]
w2 = lr.w_[0][0]
lr.fit(X_bin, y_bin, init_params=False)
assert w1 != lr.w_[0][0]
assert w2 != lr.w_[1][0]
np.testing.assert_almost_equal(lr.w_, t, 2)
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report, confusion_matrix
from mlxtend.classifier import SoftmaxRegression
from sklearn import datasets
iris = datasets.load_iris()
#X = iris.data[:, [2, 3]]
X = iris.data
y = iris.target
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.3,
random_state=0)
############# softmax Regresion ###############
# Fitting softmax regression to the tranning set
foft_regressor = SoftmaxRegression()
foft_regressor.fit(X_train, y_train)
# Predicting the test set result
y_pred = foft_regressor.predict(X_test)
print("############ softmax Regression ############")
print(confusion_matrix(y_test, y_pred))
print(classification_report(y_test, y_pred))
y = data[:, data.shape[1] - 1] # Label - shape: 150, 1
X = data[:, 0:data.shape[1] - 1].astype(float) # Data - shape: 150, 4
X_train = X[0:105, :] #shape: 120, 4
X_test = X[105:X.shape[0], :] #30, 4
y_train = y[0:105] #shape: 120, 4
y_test = y[105:y.shape[0]] #30, 4
del data, X, y
# Map label sang 0, 1, 2
classes = {'Iris-setosa': 0, 'Iris-versicolor': 1, 'Iris-virginica': 2}
y_train = [classes[item] for item in y_train]
y_test = [classes[item] for item in y_test]
y_train = np.asarray(y_train)
y_test = np.asarray(y_test)
# Softmax
softmax = SoftmaxRegression(eta=1 / (10 ^ 4),
epochs=500,
minibatches=1,
random_seed=0,
print_progress=3)
softmax.fit(X_train, y_train, init_params=True)
"""
plt.plot(range(len(softmax.cost_)), softmax.cost_)
plt.xlabel('Iterations')
plt.ylabel('Cost')
plt.show()
"""
accuracy = softmax.score(X_test, y_test)
print(accuracy)
def main():
#SETUP!!!
train = 0.9 #percetage of data for training
dev = 0.05 #percetage of data for development
test = 0.05 #percetage of data for test
n_features = 1500 #this could be adjusted later by the algorithm
#this is setting the CountVectorizer from sklearn.feature_extraction.text
vectorizer = CountVectorizer(
min_df=20, #you may want to adjust this
max_features=n_features,
lowercase=False)
DO_STANDARDIZE_DATA = 1 #1 yes, 0 no
regularization_lambda = 0.1
ETA = 0.00005
EPOCHS = 50
model_sm = SoftmaxRegression(
eta=ETA,
epochs=EPOCHS,
l2=regularization_lambda,
#n_classes=U,
minibatches=1,
random_seed=1,
print_progress=3)
print("-----------------------------")
print("METHOD - SOFTMAX REGRESSION")
print("-----------------------------")
print("Hello,\nwe will use Softmax Regression to classify twitter users\n")
setpath()
#get the users
screen_names = get_users(FILE_USERS)
info_data = get_info()
U = len(screen_names) #number of users
for i in range(U):
print("For", screen_names[i], " one has ", info_data[i, 1], "tweets")
if os.path.isfile(FOLDER + "/update_SM" + str(U) + ".txt") == True:
with open(FOLDER + "/update_SM" + str(U) + ".txt", "r") as h:
update = h.read()
h.close()
print("We load the dataset.")
file = FOLDER + "/X_train_politic" + update + ".npy"
with open(file, 'rb') as f:
X_train = pickle.load(f)
file = FOLDER + "/Y_train_politic" + update + ".npy"
with open(file, 'rb') as f:
Y_train = pickle.load(f)
file = FOLDER + "/X_dev_politic" + update + ".npy"
with open(file, 'rb') as f:
X_dev = pickle.load(f)
file = FOLDER + "/Y_dev_politic" + update + ".npy"
with open(file, 'rb') as f:
Y_dev = pickle.load(f)
file = FOLDER + "/X_test_politic" + update + ".npy"
with open(file, 'rb') as f:
X_test = pickle.load(f)
file = FOLDER + "/Y_test_politic" + update + ".npy"
with open(file, 'rb') as f:
Y_test = pickle.load(f)
else:
all_tweets = load_data()
random.shuffle(all_tweets)
random.shuffle(
all_tweets) #Always shuffle your opponent cards when you play :)
tweets = []
YY = []
for i in range(len(all_tweets)):
tweets.append(all_tweets[i][2])
YY.append(all_tweets[i][0])
if len(tweets) == len(all_tweets):
print("We load the data and we create the data set!")
Y = np.array(YY) #this is the output label vector
print("-----------------------------")
m = len(tweets)
X_train_1, x_appoggio, Y_train, y_appoggio = train_test_split(
tweets, Y, test_size=(dev + test))
X_dev_1, X_test_1, Y_dev, Y_test = train_test_split(
x_appoggio, y_appoggio, test_size=(test / (dev + test)))
print("We will train with the", train * 100, " % of the data;")
print(dev * 100,
"% of the data is reserve for the method development;")
print(test * 100, "% of the data is for the test.")
vectorizer.fit(X_train_1)
X_train = vectorizer.transform(X_train_1)
X_dev = vectorizer.transform(X_dev_1)
X_test = vectorizer.transform(X_test_1)
if DO_STANDARDIZE_DATA == 0:
print("We don't standardize data")
else:
print(
"We will provide to the model with standardize data, mean zero and variance 1"
)
X_train, X_dev, X_test = standardize_data(X_train, X_dev, X_test)
del (all_tweets)
del (X_train_1, X_dev_1, X_test_1, x_appoggio, y_appoggio)
today = date.today()
today_string = today.strftime("%y_%b_%d")
#we save the data we have prepared
with open(
FOLDER + "/X_train_politic" + today_string + "_SM" + str(U) +
".npy", 'wb') as f:
pickle.dump(X_train, f)
file = FOLDER + "/X_train_politic" + today_string + "_SM" + str(
U) + ".npy"
with open(file, "wb") as f:
pickle.dump(X_train, f)
file = FOLDER + "/Y_train_politic" + today_string + "_SM" + str(
U) + ".npy"
with open(file, "wb") as f:
pickle.dump(Y_train, f)
file = FOLDER + "/X_dev_politic" + today_string + "_SM" + str(
U) + ".npy"
with open(file, "wb") as f:
pickle.dump(X_dev, f)
file = FOLDER + "/Y_dev_politic" + today_string + "_SM" + str(
U) + ".npy"
with open(file, "wb") as f:
pickle.dump(Y_dev, f)
file = FOLDER + "/X_test_politic" + today_string + "_SM" + str(
U) + ".npy"
with open(file, "wb") as f:
pickle.dump(X_test, f)
file = FOLDER + "/Y_test_politic" + today_string + "_SM" + str(
U) + ".npy"
with open(file, "wb") as f:
pickle.dump(Y_test, f)
with open(FOLDER + "/update_SM" + str(U) + ".txt", "w") as h:
h.write(today_string + "_SM" + str(U))
h.close()
D = X_test.toarray().shape[1] #this is the lengh of the input vector
print("\n")
if n_features > D:
n_features = D
print("The # of features is", n_features)
print("The regularization parameter is", regularization_lambda)
print("The learning step is", ETA)
print("The # of cycle is", EPOCHS)
print("\n")
#WE START TRAINING THE MODEL
model_sm.fit(X_train.toarray(), Y_train)
acc = model_sm.score(X_train.toarray(), Y_train)
acc_dev = model_sm.score(X_dev.toarray(), Y_dev)
print("\n")
print("Accuracy on the training set", acc)
print("Accuracy on the development set", acc_dev)
#print some statistics about the model
df_score, df_fp, df_pre = compute_accuracies(model_sm, 1, screen_names,
X_train, X_dev, Y_train,
Y_dev)
from mlxtend.data import iris_data
from mlxtend.plotting import plot_decision_regions
from mlxtend.classifier import SoftmaxRegression
import matplotlib.pyplot as plt
# Loading Data
X, y = iris_data()
X = X[:, [0, 3]] # sepal length and petal width
# standardize
X[:, 0] = (X[:, 0] - X[:, 0].mean()) / X[:, 0].std()
X[:, 1] = (X[:, 1] - X[:, 1].mean()) / X[:, 1].std()
lr = SoftmaxRegression(eta=0.01,
epochs=500,
minibatches=1,
random_seed=1,
print_progress=3)
lr.fit(X, y)
plot_decision_regions(X, y, clf=lr)
plt.title('Softmax Regression - Gradient Descent')
plt.show()
plt.plot(range(len(lr.cost_)), lr.cost_)
plt.xlabel('Iterations')
plt.ylabel('Cost')
plt.show()
def test_multi_logistic_regression_gd_weights():
t = np.array([[0.58, -3.72, 3.15], [-3.52, 3.21, 0.28]])
lr = SoftmaxRegression(epochs=200, eta=0.005, minibatches=1, random_seed=1)
lr.fit(X, y)
np.testing.assert_almost_equal(lr.w_, t, 2)
