parser = argparse.ArgumentParser(
description='T-test to see if TAN and naive are different')
parser.add_argument('-data', type=str, help='Data set path')
args = parser.parse_args()
data = args.data
data = data.replace('\r', '') #Removes the carriage return cuz I use windows
#Load the data in
with open(data, "r") as read_file:
data = json.load(read_file)
metadata = np.array(data['metadata']['features'])
#Build a list of list containing accuracy for naive and TAN
accuracy = []
for i in range(0, 10):
split = split_data(data, i, 10)
train = split[0]
test = split[1]
accuracy.append(ttest(train, test, metadata))
accuracy = np.array(accuracy)
differences = accuracy.T[1] - accuracy.T[0]
print("Accuracy (NB, TAN): \n", accuracy)
print("Differences (TAN - NB): \n", differences)
#Calculate average accuracy difference:
diff_avg = np.mean(differences)
#Calculate standard deviation of differences
diff_sd = np.std(differences)
#Calculate standard error from SD
method = ['DBSCAN']
#driver_main, impostor, n_clusters, selected_features, window_size, method = fc.general_parameters()
n_estimators = fc.if_parameters()
kernel, nu = fc.ocsvm_parameters()
for driver in driver_main:
for ws in window_size:
print("Building DF for Driver", driver, "with Window_Size", ws)
data_final = fc.build_df_final(data_normalized, driver, ws, selected_features)
print('Bulding DF with all impostors')
data_impostor, impostores = fc.build_impostors_df(data_normalized, impostor, ws, selected_features, driver)
print('Doing data split')
x_train, x_val = fc.split_data(data_final)
for c in n_clusters:
print('Create clusters')
labels_train, centroid_train, x_train_class = fc.clusters_of_maneuvers(x_train, c)
for m in method:
if (m=='DBSCAN'):
print('Training DBSCAN')
dbscan_list, eps_list = fc.train_model_dbscan(labels_train, centroid_train, x_train_class, x_val)
print('Doing predictions DBSCAN')
result = fc.test_model_dbscan(dbscan_list, data_final, data_impostor, centroid_train, x_train_class)
print('Evaluanting the results')
acc, min_man, media_man, max_man, deviation = fc.evaluating_result(result, ws)
from keras.layers import Dropout
from keras.models import model_from_json
from keras.models import load_model
from nltk.tokenize import RegexpTokenizer
path = '/home/mark/Research'
data_dir = path + '/data'
train = True
load_all = True
weight_matrix, word_index = functions.load_embeddings(data_dir +
'/glove.6B.100d.txt')
data = functions.read_data(data_dir)
train, test, val = functions.split_data(data, .8, data_dir)
train = train.reset_index()
test = test.reset_index()
val = val.reset_index()
#max_length, avg_words, seq_length = functions.maxLen(data)
train_x = functions.pipeline(train, word_index, weight_matrix)
test_x = functions.pipeline(test, word_index, weight_matrix)
val_x = functions.pipeline(val, word_index, weight_matrix)
train_y = functions.labels(train)
test_y = functions.labels(test)
val_y = functions.labels(val)
print 'Training data: '
# get data
import functions as f
#import cv2
from sklearn.neighbors import KNeighborsClassifier
import numpy as np
from sklearn import datasets, svm, metrics
data = f.get_array_from_images('images_no_copies')
dev, test = f.split_data(data, 0.2)
training_data, training_labels = f.reshape(dev)
test_data, test_labels = f.reshape(test)
classifier = svm.SVC(gamma=0.001)
classifier.fit(training_data, training_labels)
score = classifier.score(test_data, test_labels)
print(score)
"Вас вітає мережа магазинів WINTIME, якщо вы хочете дізнатися інформацію про магазин в вашому місті - введіть назву свого міста"
)
print(
"Команди: /all - подивитися доступні міста, /add - додати місто, /edit - редагувати місто, /delete - видалити місто"
)
print("==" * 20)
while True:
user_input = input("Ввести дані ( для виходу - exit ): ")
if user_input != "exit" and user_input not in options:
get = wc.getInfoCities(user_input)
elif user_input.strip() == "/add":
next_input = input(
"Ведіть назву міста, адресу, телефон та години роботи через кому (city,address,phone,time): "
)
new_city = f.split_data(next_input)
if len(new_city) < 4:
print("Ви ввели недостатньо даних, спробуйте ще раз:(")
else:
add = wc.addCity(new_city)
elif user_input.strip() == "/edit":
# print("Введіть назву міста, яке ви хочете редагувати\nДоступні міста")
# cities = wc.getCities()
edit_city = input(
"Введіть назву міста та НОВІ ДАНІ через кому (city, new address, new phone, new time):"
)
upd_city = f.split_data(edit_city)
print(upd_city)
edit = wc.editCity(upd_city)
logging.config.dictConfig(config)
logger = logging.getLogger('base')
# 2
logger.info("setting metaparameters")
n_epochs = 10
learning_rate = 0.01
batch_size = 100
logger.info("n_epochs: {}, learning_rate: {}, batch_size: {}"
.format(n_epochs, learning_rate, batch_size))
# 3
logger.info("data preparation")
X_np, y_np = functions.get_data()
m, n = X_np.shape
X_split, y_split, n_batches = functions.split_data(X_np, y_np, batch_size, m)
# 4
logger.info("starting construction phase")
X = tf.placeholder(tf.float32, shape=(None, n), name="X")
y = tf.placeholder(tf.float32, shape=(None, 1), name="y")
theta = tf.Variable(tf.random_uniform([n, 1], -1.0, 1.0), name="theta")
y_pred = tf.matmul(X, theta, name="predictions")
# 5
with tf.name_scope("loss") as scope:
error = y_pred - y
mse = tf.reduce_mean(tf.square(error), name="mse")
def main():
st.title("Machine Learning Binary Classification Web App")
st.markdown("Classifying the mushrooms 🍄 to be Edible or Poisonous")
st.markdown("(Configure classifier options on the left side menu)")
st.sidebar.title("Machine Learning Binary Classification Web App")
st.sidebar.markdown(
"Classifying the mushrooms 🍄 to be Edible or Poisonous")
df = load_data()
if st.checkbox("Show raw data", True):
st.subheader("Mushroom Data Set (Classification")
st.write(df)
x_train, x_test, y_train, y_test = split_data(df)
class_names = ['edible', 'poisonous']
st.sidebar.subheader("Choose Classifier")
classifier = st.sidebar.selectbox(
"Classifier", ("Support Vector Machine (SVM)", "Logistic Regression", "Random Forest"))
if classifier == "Support Vector Machine (SVM)":
st.sidebar.subheader("Model Hyperparameters")
C = st.sidebar.number_input(
"C (Regularisation parameter)", 0.01, 10.0, step=0.01, key="C")
kernel = st.sidebar.radio("Kernel", ("rbf", "linear"), key="kernel")
gamma = st.sidebar.radio(
"Gamma (Kernal Coefficient)", ("scale", "auto"), key="gamma")
metrics = st.sidebar.multiselect("What metrics to plot?", (
'Confusion Matrix', 'ROC Curve', 'Precision-Recall Curve'), key='metrics')
if st.sidebar.button("Classify", key='classify'):
st.subheader("Support Vector Machine (SVM) Results: ")
model = SVC(C=C, kernel=kernel, gamma=gamma)
model.fit(x_train, y_train)
accuracy = model.score(x_test, y_test)
y_pred = model.predict(x_test)
st.write("Accuracy: ", accuracy.round(2))
st.write("Precision: ", precision_score(
y_test, y_pred, labels=class_names).round(2))
st.write("Recall: ", recall_score(
y_test, y_pred, labels=class_names).round(2))
plot_metrics(metrics, model, x_test, y_test, class_names)
if classifier == "Logistic Regression":
st.sidebar.subheader("Model Hyperparameters")
C = st.sidebar.number_input(
"C (Regularisation parameter)", 0.01, 10.0, step=0.01, key="C_LR")
max_iter = st.sidebar.slider(
"Maximum number of iterations", 100, 500, key='max_iter')
metrics = st.sidebar.multiselect("What metrics to plot?", (
'Confusion Matrix', 'ROC Curve', 'Precision-Recall Curve'), key='metrics')
if st.sidebar.button("Classify", key='classify'):
st.subheader("Logistic Regression Results: ")
model = LogisticRegression(C=C, max_iter=max_iter)
model.fit(x_train, y_train)
accuracy = model.score(x_test, y_test)
y_pred = model.predict(x_test)
st.write("Accuracy: ", accuracy.round(2))
st.write("Precision: ", precision_score(
y_test, y_pred, labels=class_names).round(2))
st.write("Recall: ", recall_score(
y_test, y_pred, labels=class_names).round(2))
plot_metrics(metrics, model, x_test, y_test, class_names)
if classifier == "Random Forest":
st.sidebar.subheader("Model Hyperparameters")
n_estimators = st.sidebar.number_input(
"The number of trees in the forest", 100, 5000, step=10, key='n_estimators')
max_depth = st.sidebar.number_input(
"The maximum depth of the tree", 1, 20, step=1, key='max_depth')
bootstrap = st.sidebar.radio(
"Bootstrap samples when building trees", ('True', 'False'), key='bootstrap')
metrics = st.sidebar.multiselect("What metrics to plot?", (
'Confusion Matrix', 'ROC Curve', 'Precision-Recall Curve'), key='metrics')
if st.sidebar.button("Classify", key='classify'):
st.subheader("Random Forest Results: ")
model = RandomForestClassifier(
n_estimators=n_estimators, max_depth=max_depth, bootstrap=bootstrap, n_jobs=-1)
model.fit(x_train, y_train)
accuracy = model.score(x_test, y_test)
y_pred = model.predict(x_test)
st.write("Accuracy: ", accuracy.round(2))
st.write("Precision: ", precision_score(
y_test, y_pred, labels=class_names).round(2))
st.write("Recall: ", recall_score(
y_test, y_pred, labels=class_names).round(2))
plot_metrics(metrics, model, x_test, y_test, class_names)
# get data from functions import get_all_data, split_data, reshape import cv2 from sklearn.neighbors import KNeighborsClassifier import numpy as np from sklearn import datasets, svm, metrics import os import cv2 import pickle data = get_all_data() dev, test = split_data(data, 0.2) training_data, training_labels = reshape(dev) test_data, test_labels = reshape(test) classifier = svm.SVC(gamma=0.001) classifier.fit(training_data, training_labels) # save the model to disk filename = 'model.sav' pickle.dump(classifier, open(filename, 'wb')) #score = classifier.score(test_data, test_labels) #print(score)
def run_regression(X,
z,
reg_string,
polydegree,
lambdas,
N_bs,
K,
test_size,
scale,
max_iter=50000):
"""
Runs the selected regression methods for the input design matrix, p's, lambdas, and using
the resampling methods as specified.
While there may be several ways I could have done this more optimally, this function exists
because a rather late attempt at restructuring the code in order to reduce the amount of duplicate
lines of code regarding regression, that had just escalated out of control, making it extremely
difficult to debug and finding whatever was causing all the issues.
:param X: (N, p) array containing input design matrix
:param z: (N, 1) array containing data points
:param reg_string: string containing the name of the regression method to be used
:param polydegree: list/range of the different p-values to be used
:param lambdas: array of all the lambda values to be used
:param N_bs: int, number of Bootstraps
:param K: int, number of folds in the Cross-Validation
:param test_size: float, size of the test partition [0.0, 1.0]
:param scale: list determining if the scaling is only by the mean, the std or both [bool(mean), bool(std)]
:param max_iter: maximum number of iterations for Lasso
:return: a lot of arrays with the various results and different ways of representing the data
"""
nlambdas = len(lambdas) # number of lambdas
p = polydegree[-1] # the maximum p-value
method = 4 # OLS method
# Splitting into train and test, scaling the data
X_train, X_test, z_train, z_test = fun.split_data(X,
z,
test_size=test_size)
X_train_scaled = fun.scale_X(X_train, scale)
X_test_scaled = fun.scale_X(X_test, scale)
X_scaled = fun.scale_X(X, scale)
# Bootstrap arrays
bs_error_train = np.zeros((p, nlambdas))
bs_error_test = np.zeros((p, nlambdas))
bs_bias = np.zeros((p, nlambdas))
bs_var = np.zeros((p, nlambdas))
bs_error_train_opt = np.zeros((p, 2))
bs_error_test_opt = np.zeros((p, 2))
bs_bias_opt = np.zeros(
(p, 2)
) # First index is min(MSE) lmb for each p, second at lmb that yields total lowest MSE
bs_var_opt = np.zeros((p, 2))
bs_lmb_opt = np.zeros(p)
# Cross-validation arrays
cv_error_train = np.zeros((p, nlambdas))
cv_error_test = np.zeros((p, nlambdas))
cv_error_train_opt = np.zeros((p, 2))
cv_error_test_opt = np.zeros((p, 2))
cv_lmb_opt = np.zeros(p)
# Setting up regression object to be used for regression (Lasso is dealt with later)
reg_obj = reg.OrdinaryLeastSquares(method) # default
if reg_string == 'SKL':
reg_obj = skl.LinearRegression() # Testing with scikit-learn OLS
elif reg_string == 'Ridge':
reg_obj = reg.RidgeRegression()
# Looping over all polynomial degrees in the analysis
for degree in polydegree:
n_poly = fun.polynom_N_terms(
degree
) # number of terms in the design matrix for the given degree
print('p = %2d, np = %3d' % (degree, n_poly))
# Setting up correct design matrices for the current degree
X_train_bs = np.zeros((len(X_train_scaled), n_poly))
X_test_bs = np.zeros((len(X_test_scaled), n_poly))
X_cv = np.zeros((len(X_scaled), n_poly))
# Filling the elements up to term n_poly
X_train_bs[:, :] = X_train_scaled[:, 0:n_poly]
X_test_bs[:, :] = X_test_scaled[:, 0:n_poly]
X_cv[:, :] = X_scaled[:, 0:n_poly]
# Looping over all the lambda values
for i in range(nlambdas):
lmb = lambdas[i] # current lambda value
# Printing out in order to gauge where we are
if i % 10 == 0:
print('i = %d, lmb= %.3e' % (i, lmb))
# Updating the current lambda value for Ridge and Lasso
if reg_string == 'Ridge':
reg_obj.set_lambda(lmb)
elif reg_string == 'Lasso':
reg_obj = skl.Lasso(alpha=lmb,
max_iter=max_iter,
precompute=True,
warm_start=True)
# Bootstrap
BS = res.Bootstrap(X_train_bs, X_test_bs, z_train, z_test, reg_obj)
error_, bias_, var_, trainE_ = BS.compute(
N_bs) # performing the Bootstrap
bs_error_test[degree - 1, i] = error_
bs_bias[degree - 1, i] = bias_
bs_var[degree - 1, i] = var_
bs_error_train[degree - 1, i] = trainE_
# Cross validation
CV = res.CrossValidation(X_cv, z, reg_obj)
trainE, testE = CV.compute(K) # performing the Cross-Validation
cv_error_train[degree - 1, i] = trainE
cv_error_test[degree - 1, i] = testE
# Locating minimum MSE for each polynomial degree
# Bootstrap
index_bs = np.argmin(bs_error_test[degree - 1, :])
bs_lmb_opt[degree - 1] = lambdas[index_bs]
bs_error_train_opt[:, 0] = bs_error_train[:, index_bs]
bs_error_test_opt[:, 0] = bs_error_test[:, index_bs]
bs_bias_opt[:, 0] = bs_bias[:, index_bs]
bs_var_opt[:, 0] = bs_var[:, index_bs]
# Cross-validation
index_cv = np.argmin(cv_error_test[degree - 1, :])
cv_lmb_opt[degree - 1] = lambdas[index_cv]
cv_error_train_opt[:, 0] = cv_error_train[:, index_cv]
cv_error_test_opt[:, 0] = cv_error_test[:, index_cv]
# Locate minimum MSE to see how it depends on lambda
bs_min = np.unravel_index(np.argmin(bs_error_test), bs_error_test.shape)
cv_min = np.unravel_index(np.argmin(cv_error_test), cv_error_test.shape)
bs_best = [polydegree[bs_min[0]], lambdas[bs_min[1]]]
cv_best = [polydegree[cv_min[0]], lambdas[cv_min[1]]]
# Bootstrap
bs_error_train_opt[:, 1] = bs_error_train[:, bs_min[1]]
bs_error_test_opt[:, 1] = bs_error_test[:, bs_min[1]]
bs_bias_opt[:, 1] = bs_bias[:, bs_min[1]]
bs_var_opt[:, 1] = bs_var[:, bs_min[1]]
# Cross-validation
cv_error_train_opt[:, 1] = cv_error_train[:, cv_min[1]]
cv_error_test_opt[:, 1] = cv_error_test[:, cv_min[1]]
# This return is extremely large, sadly, and should have been improved upon
# this was just the fastest way of doing it when I had to restructure the code
# so better planning in the future would be a better solution
return (bs_error_train, bs_error_test, bs_bias, bs_var, bs_error_train_opt,
bs_error_test_opt, bs_bias_opt, bs_var_opt, bs_lmb_opt,
cv_error_train, cv_error_test, cv_error_train_opt,
cv_error_test_opt, cv_lmb_opt, bs_min, bs_best, cv_min, cv_best)
# This return is extremely large, sadly, and should have been improved upon
# this was just the fastest way of doing it when I had to restructure the code
# so better planning in the future would be a better solution
return (bs_error_train, bs_error_test, bs_bias, bs_var, bs_error_train_opt,
bs_error_test_opt, bs_bias_opt, bs_var_opt, bs_lmb_opt,
cv_error_train, cv_error_test, cv_error_train_opt,
cv_error_test_opt, cv_lmb_opt, bs_min, bs_best, cv_min, cv_best)
########################################################################################################################
if run_mode == 'a':
save = 'N%d_nf%d_p%d_noise%.2f_seed%d' % (N, n_franke, p, noise, seed)
# Splitting into train and test data
X_train, X_test, z_train, z_test = fun.split_data(X,
z_ravel,
test_size=test_size)
# X_train, X_test, z_train, z_test = train_test_split(X, z_ravel, test_size=test_size)
# Scaling the data
X_train_scaled = fun.scale_X(X_train, scale)
X_test_scaled = fun.scale_X(X_test, scale)
# Plotting the Franke function
fun.plot_surf(x_mesh,
y_mesh,
z_mesh,
'x',
'y',
'z',
'Franke function, $N$=%d, noise=%.2f' % (N, noise),