def main():
arr = utils.get_array()
len_arr = len(arr)
for j in range(1, len_arr):
i = j - 1
key = arr[j]
while i >= 0 and arr[i] > key:
arr[i + 1] = arr[i]
i = i - 1
arr[i + 1] = key
print(arr)
def __init__(self, path, transform=None):
self.path = path
self.transform = transform
self.array = utils.get_array(path)
import utils
def merge(left, right):
A = []
i ,j = 0, 0
while i < len(left) and j < len(right):
if left[i] <= right[j]:
A.append(left[i])
i += 1
else:
A.append(right[j])
j += 1
A += left[i:]
A += right[j:]
return A
def mergesort(A):
if len(A) > 1:
q = len(A) // 2
left = mergesort(A[:q])
right = mergesort(A[q:])
return merge(left, right)
return A
if __name__ == "__main__":
A = utils.get_array()
print "unsorted(A): " + str(A)
print "mergesort(A): " + str(mergesort(A))
def get_physical_chan_bool_vtr_attribute(devName, attrId): values = utils.get_array(c_daqmx.DAQmxGetPhysicalChanAttribute, ctypes.c_uint32, ctypes.c_char_p(devName), ctypes.c_int32(attrId)) return [bool(b) for b in values]
def get_physical_chan_double_vtr_attribute(devName, attrId): return utils.get_array(c_daqmx.DAQmxGetPhysicalChanAttribute, ctypes.c_double, devName, ctypes.c_int32(attrId))
def get_device_bool_vtr_attribute(devName, attrId): values = utils.get_array(c_daqmx.DAQmxGetDeviceAttribute, ctypes.c_uint32, ctypes.c_char_p(devName), ctypes.c_int32(attrId)) return [bool(b) for b in values]
def get_device_double_vtr_attribute(devName, attrId): return utils.get_array(c_daqmx.DAQmxGetDeviceAttribute, ctypes.c_double, devName, ctypes.c_int32(attrId))
#! /usr/bin/python
# coding: utf-8
import utils
def insertion_sort(A):
for i in range(1, len(A)):
key = A[i]
j = i - 1
while j >= 0 and A[j] > key:
A[j + 1] = A[j]
j -= 1
A[j + 1] = key
if __name__ == "__main__":
A = utils.get_array()
print "unsorted(A): " + str(A)
insertion_sort(A)
print "sorted(A): " + str(A)
def train(X ,y, groups, algo_option, feature_option, balancing_option, scale_option, reduce_dimension_option):
# Read processed file
X_subset = get_subset_features(X, feature_option)
y_subset = deepcopy(y)
logo = StratifiedShuffleSplit(n_splits=50, test_size=0.2, random_state=0)
fold_accuracy_scores = np.zeros(0)
fold_f1_macro_scores = np.zeros(0)
fold_f1_weighted_scores = np.zeros(0)
fold_recall_scores = []
fold_precision_scores = []
# 5 folds corresponding to 5 events
for train_index, test_index in logo.split(X_subset, y_subset):
# Split train and test from folds
X_train, X_test = get_array(train_index, X_subset), get_array(test_index, X_subset)
y_train, y_test = get_array(train_index, y_subset), get_array(test_index, y_subset)
# Init a classifer
model = init_model(algo_option)
# Init an optional scaler
if scale_option:
scaler = init_scaler(scale_option)
scaler.fit(X_train)
X_train = scaler.transform(X_train)
X_test = scaler.transform(X_test)
# Init an optional balancing model
if balancing_option:
balancer = init_balancing_model(balancing_option)
X_train, y_train = balancer.fit_sample(X_train, y_train)
# Init an optional reduce dimenstion model
if reduce_dimension_option:
reducer = init_reduce_dimension_model(reduce_dimension_option)
reducer.fit(X_train, y_train)
X_train = reducer.transform(X_train)
X_test = reducer.transform(X_test)
# Fit prerocessed data to classifer model
model.fit(X_train, y_train)
# Predict
y_pred = model.predict(X_test)
# Metrics
matrix = confusion_matrix(np.asarray(y_test), y_pred)
# false_false_rate = 1.0* matrix[0][1] / sum(matrix[0]) # could be high
# false_true_rate = 1.0* matrix[1][0] / sum(matrix[:, 0]) # must be low
current_fold_accuracy = f1_score(np.asarray(y_test), y_pred, average='micro')
current_fold_macro_f1 = f1_score(np.asarray(y_test), y_pred, average='macro')
current_fold_weighted_f1 = f1_score(np.asarray(y_test), y_pred, average='weighted')
current_recall = recall_score(np.asarray(y_test), y_pred, average=None)
current_precision = precision_score(np.asarray(y_test), y_pred, average=None)
# print "Micro f1-score (Accuracy):\t\t\t", current_fold_accuracy
# print "Macro f1-score:\t\t\t", current_fold_macro_f1
# print "Weighted f1-score:\t\t\t", current_fold_weighted_f1
# print "Rate false of false label:\t\t\t", false_false_rate
# print "Rate false of true label:\t\t\t", false_true_rate
fold_accuracy_scores = np.append(fold_accuracy_scores,current_fold_accuracy)
fold_f1_macro_scores = np.append(fold_f1_macro_scores, current_fold_macro_f1)
fold_f1_weighted_scores = np.append(fold_f1_weighted_scores, current_fold_weighted_f1)
fold_recall_scores.append(current_recall)
fold_precision_scores.append(current_precision)
# print current_recall
# print current_precision
# print confusion_matrix(np.asarray(y_test), y_pred)
# tmp = []
# for (index,x) in enumerate(model.feature_importances_):
# if x!=0:
# tmp.append((x,index))
# print sorted(tmp, reverse=True)
# raw_input()
# print "Accuracy:\t\t", fold_accuracy_scores, '\t\t', fold_accuracy_scores.mean()
# print "F1-macro:\t\t", fold_f1_macro_scores, '\t\t', fold_f1_macro_scores.mean()
# print "F1-weighted:\t", fold_f1_weighted_scores, '\t\t', fold_f1_weighted_scores.mean()
print "Accuracy:\t\t", fold_accuracy_scores.mean()
print "F1-macro:\t\t", fold_f1_macro_scores.mean()
print "F1-weighted:\t", fold_f1_weighted_scores.mean()
print "Recall: \t\t", np.asarray(fold_recall_scores).mean(axis=0)
print "Precision: \t\t", np.asarray(fold_precision_scores).mean(axis=0)
# TRAIN AND SAVE A MODEL FOR TESTING ON SEMEVAL TEST SET
X_train = X_subset
y_train = y_subset
# Init a classifer
model = init_model(algo_option)
# Init an optional scaler
scaler = None
if scale_option:
scaler = init_scaler(scale_option)
scaler.fit(X_train)
X_train = scaler.transform(X_train)
# Init an optional balancing model
balancer = None
if balancing_option:
balancer = init_balancing_model(balancing_option)
X_train, y_train = balancer.fit_sample(X_train, y_train)
# Init an optional reduce dimenstion model
reducer = None
if reduce_dimension_option:
reducer = init_reduce_dimension_model(reduce_dimension_option)
reducer.fit(X_train, y_train)
X_train = reducer.transform(X_train)
# Fit prerocessed data to classifer model
model.fit(X_train, y_train)
# Save model
pickle.dump(model, open(os.path.join(MODELS_ROOT,'classifier.model'),"wb"))
if os.path.exists(os.path.join(MODELS_ROOT, 'scaler.model')):
os.remove(os.path.join(MODELS_ROOT, 'scaler.model'))
if scaler != None:
pickle.dump(scaler, open(os.path.join(MODELS_ROOT, 'scaler.model'), "wb"))
if os.path.exists(os.path.join(MODELS_ROOT, 'balancer.model')):
os.remove(os.path.join(MODELS_ROOT, 'balancer.model'))
if balancer != None:
pickle.dump(balancer, open(os.path.join(MODELS_ROOT, 'balancer.model'), "wb"))
if os.path.exists(os.path.join(MODELS_ROOT, 'reducer.model')):
os.remove(os.path.join(MODELS_ROOT, 'reducer.model'))
if reducer != None:
pickle.dump(reducer, open(os.path.join(MODELS_ROOT, 'reducer.model'), "wb"))
training_settings = {
'features_subset': feature_option,
'balancing_class_algorithm': balancing_option,
'scale_option': scale_option,
'reduce_dimension_algorithm': reduce_dimension_option,
'training_algorithm': algo_option
}
pickle.dump(training_settings, open(os.path.join(MODELS_ROOT,'settings.model'),"wb"))
from sklearn.metrics import mean_absolute_error
from utils import get_array, get_data, get_model_3
train, test, max_user, max_work, mapping_work = get_data('ml-20m/ratings.csv', nrows=100000)
# Train data analysis
# print(train.groupby('userId').aggregate({'movieId': 'count', 'rating': 'median', 'timestamp': ['min', 'max']}))
print(train.describe())
print((max_work, max_user))
np.random.seed(1)
model = get_model_3(max_work=max_work, max_user=max_user, latent_factors=50)
print(model.summary())
print(max(get_array(train["userId"])))
history = model.fit([get_array(train["movieId"]),
get_array(train["userId"])],
get_array(train["rating"]),
epochs=10,
# batch_size=13,
validation_split=0.2,
verbose=1)
model.save('model.h5')
predictions = model.predict([get_array(test["movieId"]), get_array(test["userId"])])
test_performance = mean_absolute_error(test["rating"], predictions)
print(" Test Mae model 3 : %s " % test_performance)
from utils import get_model_3, get_array, get_data
train, test, max_user, max_work, mapping_work = get_data('ml-20m/ratings.csv',
nrows=100000)
# Train data analysis
# print(train.groupby('userId').aggregate({'movieId': 'count', 'rating': 'median', 'timestamp': ['min', 'max']}))
print(train.describe())
print((max_work, max_user))
np.random.seed(1)
model = get_model_3(max_work=max_work, max_user=max_user, latent_factors=50)
print(model.summary())
print(max(get_array(train["userId"])))
history = model.fit(
[get_array(train["movieId"]),
get_array(train["userId"])],
get_array(train["rating"]),
epochs=10,
# batch_size=13,
validation_split=0.2,
verbose=1)
model.save('model.h5')
predictions = model.predict(
[get_array(test["movieId"]),
get_array(test["userId"])])
test_performance = mean_absolute_error(test["rating"], predictions)
