if __name__ == "__main__": model_dir = os.path.join(scene_test_list[0], 'model') result_dir = os.path.join(scene_test_list[0], 'result', 'test_out') errorlog_dir = os.path.join(scene_test_list[0], 'errorlog') summarylog_dir = os.path.join(scene_test_list[0], 'summarylog') os.makedirs(model_dir, exist_ok=True) os.makedirs(result_dir, exist_ok=True) os.makedirs(errorlog_dir, exist_ok=True) os.makedirs(summarylog_dir, exist_ok=True) train_data = dataLoader(data_dir=data_dir, subset='train', patch_width=patch_width, patch_height=patch_height, image_start_idx=train_start_idx, img_per_scene=train_per_scene, patch_per_img=patch_per_img, scene_list=scene_train_list) valid_data = dataLoader(data_dir=data_dir, subset='valid', patch_width=patch_width, patch_height=patch_height, image_start_idx=valid_start_idx, img_per_scene=valid_per_scene, patch_per_img=patch_per_img, scene_list=scene_valid_list) # Train train_dataset = tf.data.TFRecordDataset([train_data.dataset_name]) # Parse the record into tensors. train_dataset = train_dataset.map(_parse_function)
"""
Predictions are made here.
@AugustSemrau
"""
from data_loader import dataLoader
from models import Models
from data_loader import csv_saver
import xgboost as xgb
if __name__ == '__main__':
# Import test data
testData = dataLoader(test=True, ageNAN='median')
# Initiate models
Models = Models(agenan='median')
# Logistic Regression model predictions
logisticModel = Models.build_model_LR()
logistic_predictions = logisticModel.predict(testData)
# Saving predictions
csv_saver(predictions=logistic_predictions, type="logistic")
# Naive Bayes model predictions
naiveBayes_model = Models.build_model_NB()
naiveBayes_predictions = naiveBayes_model.predict(testData)
# Saving predictions
csv_saver(predictions=naiveBayes_predictions, type="naiveBayes")
# Stochastic Gradient Descent model predictions
from model.anchornet import AnchorNet
# read num_anchor from command line
parser = argparse.ArgumentParser()
parser.add_argument('--anchor', help='Number of anchors to predict')
parser.add_argument('--rectify_img',
help='Whether to rectify images',
default=0)
args = parser.parse_args()
num_anchor = int(args.anchor)
print('Number of anchors to predict: {}'.format(num_anchor))
rectify_img = True if int(args.rectify_img) > 0 else False
print('Rectify image: {}'.format(rectify_img))
# load data
data_loader = dataLoader()
imgs = data_loader.loadTestingImg()
flows = data_loader.loadTestingFlow()
total_count = len(imgs)
# load model
if num_anchor > 0:
depthnet = DepthNet(data_loader.getImgShape())
depthnet.model.load_weights(
os.path.join(os.getcwd(), "checkpoints/model_depth.hdf5"))
anchornet = AnchorNet(data_loader.getImgShape(), num_anchor)
anchornet.model.load_weights(
os.path.join(os.getcwd(),
'checkpoints/model_anchor{}.hdf5'.format(num_anchor)))
# path to save results
if __name__ == "__main__":
model_dir = os.path.join(scene_test_list[0], 'model')
result_dir = os.path.join(scene_test_list[0], 'result', 'test_out')
errorlog_dir = os.path.join(scene_test_list[0], 'errorlog')
summarylog_dir = os.path.join(scene_test_list[0], 'summarylog')
os.makedirs(model_dir, exist_ok=True)
os.makedirs(result_dir, exist_ok=True)
os.makedirs(errorlog_dir, exist_ok=True)
os.makedirs(summarylog_dir, exist_ok=True)
test_data = dataLoader(data_dir=data_dir,
subset='test',
image_start_idx=0,
img_per_scene=test_per_scene,
scene_list=scene_test_list)
# Test
test_dataset = tf.data.TFRecordDataset([test_data.dataset_name])
test_dataset = test_dataset.map(_parse_function_testdata)
test_dataset = test_dataset.batch(test_batch_size)
handle_large = tf.placeholder(tf.string, shape=[])
iterator_structure_large = tf.data.Iterator.from_string_handle(
handle_large, test_dataset.output_types, test_dataset.output_shapes)
next_element_large = iterator_structure_large.get_next()
test_iterator = test_dataset.make_initializable_iterator()
# Model
This script is a low-effort tuning of K-Nearest-Neighbor model.
@AugustSemrau
"""
from data_loader import dataLoader
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import cross_val_score
import matplotlib.pyplot as plt
import pandas as pd
if __name__ == '__main__':
# Get data
X, y = dataLoader(test=False, optimize_set=False)
y = y.values.ravel()
# First make list of k's
neighbors = list(range(1, 25, 1))
# Second make empty list for scores
cval_scores = []
# Now do 10-fold cross-validation
for K in neighbors:
model_KNN = KNeighborsClassifier(n_neighbors=K)
scores = cross_val_score(model_KNN, X, y, cv=10, scoring='accuracy')
cval_scores.append(scores.mean())
# Plotting these score to se how different number of neighbors affect performance
def plot_acc(knn_scores):
pd.DataFrame({"K": [i for i in neighbors], "Accuracy": knn_scores}).set_index("K").plot.bar(rot=0)
#formatter = logging.Formatter('[%(asctime)s] [%(name)s] [%(levelname)s] %(message)s')
logging.basicConfig(filename='logfile.log', level=logging.INFO)
formatter = logging.Formatter('[%(asctime)s] [%(name)s] [%(levelname)s] %(message)s')
logger = get_logger('ML_struct')
parser = argparse.ArgumentParser(description='Character-level convolutional neural network for text classification')
parser.add_argument('--config','--c', type=str, metavar='yaml FILE', help='where to load YAML configuration')
args = parser.parse_args()
print(args)
print(args.config)
yaml_file = args.config
with open(yaml_file) as f:
cfg = yaml.load(f)
print(cfg)
train_dataset, val_dataset = Datasets() # 이건 다른 path일때는 어떻게?
train_loader, val_loader=dataLoader(train_dataset, val_dataset)
cnn = CNNClassifier()
device = torch.device("cuda")
print(device)
if torch.cuda.device_count() >1 :
print("Let's use", torch.cuda.device_count(), "GPUs!")
#cnn.to(device)
cnn = nn.DataParallel(cnn)
cnn.to(device)
else:
cnn.to(device)
#mytensor = my_tensor.to(device)
# loss
criterion = nn.CrossEntropyLoss()
# backpropagation method
def build_optimized_RF(X, y, d0, d1, d2, d3):
model = RandomForestClassifier(n_estimators=d0,
max_depth=d1,
max_features=d2,
criterion=d3,
oob_score=True,
random_state=0)
model.fit(X, y)
return model
if __name__ == '__main__':
# Defining data
X_train, X_val, y_train, y_val = dataLoader(test=False,
optimize_set=True,
ageNAN="median")
# Building baseline RF model
baselineRF = build_baseline_RF(X=X_train, y=y_train)
print('Default RF model out-of-the-bag error', baselineRF.oob_score_)
print('Default RF model test accuracy', baselineRF.score(X_val, y_val))
# Performing Bayesian optimization to optimize RF model parameters
# Acquisition functions can be MPI, EI or LCB
opt = GPyOpt.methods.BayesianOptimization(f=objective_function,
domain=domain_RF,
acquisition_type='MPI')
opt.acquisition.exploration_weight = 0.5
opt.run_optimization(max_iter=30)
domain_best = opt.X[np.argmin(opt.Y)]
def main(_):
# Import data
#mnist = input_data.read_data_sets("./input_data/", one_hot=True)
mnist=data_loader.dataLoader("./data/train.pkl","./data/test.pkl")
# Create the model
x = tf.placeholder(tf.float32, [None, 1568])
# Define loss and optimizer
y_ = tf.placeholder(tf.float32, [None, 19])
# Build the graph for the deep net
y_conv, keep_prob = deepnn(x)
with tf.name_scope('loss'):
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(labels=y_,
logits=y_conv)
cross_entropy = tf.reduce_mean(cross_entropy)
with tf.name_scope('adam_optimizer'):
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
with tf.name_scope('accuracy'):
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))
correct_prediction = tf.cast(correct_prediction, tf.float32)
accuracy = tf.reduce_mean(correct_prediction)
graph_location = tempfile.mkdtemp()
print('Saving graph to: %s' % graph_location)
train_writer = tf.summary.FileWriter(graph_location)
train_writer.add_graph(tf.get_default_graph())
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for i in range(1001):
train_image,train_label = mnist.nextBatch(50)
if i % 500 == 0:
train_accuracy = accuracy.eval(feed_dict={
x: train_image, y_: train_label, keep_prob: 1.0})
print('step % d, training accuracy %g' % (i, train_accuracy))
train_acc.append(train_accuracy)
train_step.run(feed_dict={x: train_image, y_: train_label, keep_prob: 0.5})
if i % 500 == 0:
x_inter.append(i)
test_accuracy=accuracy.eval(feed_dict={
x: mnist.test_image, y_: mnist.test_label, keep_prob: 1.0})
print('test accuracy %g' % test_accuracy)
test_acc.append(test_accuracy)
fig = plt.figure()
ax = fig.add_subplot(1, 1, 1)
ax.plot(x_inter,train_acc, 'r', x_inter, test_acc, 'b')
label = ["train", "test"]
plt.title("accuracy of train and loss")
plt.xlabel("interations")
plt.ylabel("accuracy")
plt.legend(label, loc=1, ncol=2)
# display the plot
plt.savefig("1.jpg")
"""
This script is experimenting with the XGBoost model.
@AugustSemrau
"""
from data_loader import dataLoader
from xgboost import XGBClassifier
from sklearn.metrics import accuracy_score
if __name__ == '__main__':
# Get data
X_train, X_val, y_train, y_val = dataLoader(test=False, optimize_set=True)
y_train, y_val = y_train.values.ravel(), y_val.values.ravel()
# Define baseline XGB classifier model with default parameters
defualt_xgb = XGBClassifier()
defualt_xgb.fit(X_train, y_train)
default_predictions = defualt_xgb.predict(X_val)
print('Default XGB model test accuracy',
accuracy_score(y_val, default_predictions))
# Using early_stopping_rounds to determine best n_estimators number
tuned_xgb = XGBClassifier(n_estimators=10, learning_rate=0.5)
tuned_xgb.fit(X_train,
y_train,
early_stopping_rounds=5,
eval_set=[(X_val, y_val)],
verbose=False)
tuned_params = tuned_xgb.get_params()
print('')
def __init__(self, agenan="median"):
self.X, self.y = dataLoader(test=False,
optimize_set=False,
ageNAN=agenan)
self.y = self.y.values.ravel()
def process_CNN_word_vector(self):
data_loader = dataLoader()
# Without removing stopwords seems to give better performance for CNN
vec_max_features = 6000
vectorizer = TfidfVectorizer(stop_words=None,
ngram_range=(1, 4),
max_df=0.7,
max_features=vec_max_features)
train_docs = data_loader.get_train_docs()
dev_docs = data_loader.get_dev_docs()
test_docs = data_loader.get_test_docs()
X_train = vectorizer.fit_transform(train_docs).toarray()
X_dev = vectorizer.transform(dev_docs).toarray()
X_test = vectorizer.transform(test_docs).toarray()
# print('X_train shape: ', X_train.shape)
# X_train_vec_mean = data_loader.get_X_train_glove_vec_mean()
# X_dev_vec_mean = data_loader.get_X_dev_glove_vec_mean()
# X_test_vec_mean = data_loader.get_X_test_glove_vec_mean()
# print('X_train_vec_mean shape: ', X_train_vec_mean.shape)
# X_train = np.concatenate((X_train, X_train_vec_mean), axis=1)
# X_dev = np.concatenate((X_dev, X_dev_vec_mean), axis=1)
# X_test = np.concatenate((X_test, X_test_vec_mean), axis=1)
y_train = data_loader.get_y_train()
zip_train_X = list(zip(X_train, y_train))
random.shuffle(zip_train_X)
X_train, y_train = zip(*zip_train_X)
X_train, y_train = np.array(list(X_train)), list(y_train)
y_train = keras.utils.to_categorical(y_train, num_classes=2)
y_dev = data_loader.get_y_dev()
y_dev_categorical = keras.utils.to_categorical(y_dev, num_classes=2)
print('X_train shape:', X_train.shape)
# _activations = ['tanh', 'relu', 'selu']
# _optimizers = ['sgd', 'adam']
# _batch_size = [16, 32, 64]
# params = dict(var_activation=_activations,
# var_optimizer=_optimizers,
# batch_size=_batch_size)
# tokenizer = Tokenizer()
# tokenizer.fit_on_texts(X_train)
# maxlen = 1000
# sequences_train = tokenizer.texts_to_sequences(X_train)
# sequences_train = pad_sequences(sequences_train, maxlen=maxlen)
#
# vocab_size = len(tokenizer.word_index) + 1
# embedding_size = vec_max_features
#
# input_tfidf = Input(shape=(vec_max_features,))
# input_text = Input(shape=(maxlen,))
#
# embedding = Embedding(vocab_size, embedding_size, input_length=maxlen)(input_text)
# mean_embedding = keras.layers.Lambda(lambda x: keras.backend.mean(x, axis=1))(embedding)
# concatenated = concatenate([input_tfidf, mean_embedding])
#
# dense1 = Dense(256, activation='relu')(concatenated)
# dense2 = Dense(32, activation='relu')(dense1)
# dense3 = Dense(8, activation='sigmoid')(dense2)
#
# model = Model(inputs=[input_tfidf, input_text], outputs=dense3)
# model.summary()
# model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
num_classes = 2
verbose, epochs, batch_size = 0, 30, 16
print('Building model...')
model = Sequential()
model.add(Dropout(0.25))
model.add(Dense(256, activation='relu',
activity_regularizer=l1(0.001)))
model.add(Dense(64, activation='relu', activity_regularizer=l1(0.01)))
model.add(BatchNormalization())
model.add(Dense(num_classes, activation='softmax'))
model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
print("Fitting model... ")
model.fit(X_train,
y_train,
epochs=epochs,
batch_size=batch_size,
validation_data=(X_dev, y_dev_categorical),
shuffle=True)
y_dev_pred = model.predict_classes(X_dev)
y_test_pred = model.predict_classes(X_test)
print('Precision on development set: ',
precision_score(y_dev, y_dev_pred))
print('F1-Score on development set: ', f1_score(y_dev, y_dev_pred))
print('Recall on development set: ', recall_score(y_dev, y_dev_pred))
print('Negative prediction proportion on development set: ',
sum([1 for y in y_dev_pred if y == 0]) / len(y_dev_pred))
print('Negative prediction proportion on test set: ',
sum([1 for y in y_test_pred if y == 0]) / len(y_test_pred))
return y_test_pred
def process_cnn_embedding(self):
data_loader = dataLoader()
train_docs = data_loader.get_train_docs()
y_train = data_loader.get_y_train()
zip_train_X = list(zip(train_docs, y_train))
random.shuffle(zip_train_X)
train_docs, y_train = zip(*zip_train_X)
train_docs, y_train = list(train_docs), list(y_train)
dev_docs = data_loader.get_dev_docs()
test_docs = data_loader.get_test_docs()
self.tokenizer.fit_on_texts(train_docs + dev_docs + test_docs)
encoded_docs = self.tokenizer.texts_to_sequences(train_docs)
# pad documents to a max length of 10000 words
max_length = 10000
X_train = pad_sequences(encoded_docs,
maxlen=max_length,
padding='post')
y_train = keras.utils.to_categorical(y_train, num_classes=2)
# vocab_size = len(self.tokenizer.word_index) + 1
glove_vec_dim = 300
word_index = self.tokenizer.word_index
embedding_matrix = self.create_cnn_embedding_matrix(
word_index, self.vec_dict, max_length)
emb_mean, emb_std = -0.005838499, 0.48782197
all_embs = np.stack(self.vec_dict.values())
embed_size = all_embs.shape[1]
nb_words = min(max_length, len(word_index))
embedding_matrix2 = np.random.normal(emb_mean, emb_std,
(nb_words, embed_size))
print('Embedding matrix shape: ', embedding_matrix.shape)
print('Building model...')
# input1 = Input(shape=(embedding_matrix.shape[0],))
# embedding1 = Embedding(embedding_matrix.shape[0], glove_vec_dim, weights=[embedding_matrix],
# input_length=embedding_matrix.shape[0])(input1)
# conv1 = Conv1D(filters=16, kernel_size=3, activation='relu', activity_regularizer=l1(0.001))(embedding1)
# drop1 = Dropout(0.2)(conv1)
# conv1 = MaxPooling1D(pool_size=2)(drop1)
#
# input2 = Input(shape=(embedding_matrix.shape[0],))
# embedding2 = Embedding(embedding_matrix.shape[0], glove_vec_dim,
# input_length=embedding_matrix.shape[0], trainable=True)(input2)
# conv2 = Conv1D(filters=16, kernel_size=3, activation='relu',)(embedding2)
# drop2 = Dropout(0.2)(conv2)
# conv2 = MaxPooling1D(pool_size=2)(drop2)
# input2 = Input(shape=(embedding_matrix.shape[0],))
# embedding2 = Embedding(embedding_matrix.shape[0], glove_vec_dim, trainable=True)(input2)
# conv2 = Conv1D(filters=32, kernel_size=4, activation='relu', activity_regularizer=l1(0.001))(embedding2)
# drop1 = Dropout(0.2)(conv2)
# conv2 = MaxPooling1D(pool_size=2)(drop1)
# cnn = concatenate([conv1, conv2], axis=-1)
# flat = Flatten()(cnn)
# # normal = BatchNormalization()(flat)
# # x = Dense(128, activation="relu")(flat)
# x = Dense(2, activation="softmax")(flat)
# model = Model(inputs=[input1, input2], outputs=x)
# model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
# # plot_model(model, show_shapes=True, to_file='multichannel.png')
model = Sequential()
model.add(
Embedding(embedding_matrix.shape[0],
glove_vec_dim,
weights=[embedding_matrix],
input_length=embedding_matrix.shape[0]))
model.add(Dropout(0.25))
model.add(
Conv1D(filters=4,
kernel_size=3,
activation='relu',
activity_regularizer=l1(0.01)))
# model.add(SpatialDropout1D(0.2))
model.add(MaxPooling1D(pool_size=4))
# model.add(Conv1D(filters=128, kernel_size=4, activation='relu', activity_regularizer=l1(0.001)))
# model.add(MaxPooling1D(pool_size=4))
# model.add(Conv1D(filters=128, kernel_size=4, activation='relu', activity_regularizer=l1(0.001)))
# model.add(MaxPooling1D(pool_size=4))
# model.add(LSTM(70))
# model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(BatchNormalization())
model.add(Dense(2, activation='softmax'))
model.compile(optimizer='adam',
loss='binary_crossentropy',
metrics=['accuracy'])
print(model.summary())
encoded_dev_docs = self.tokenizer.texts_to_sequences(dev_docs)
X_dev = pad_sequences(encoded_dev_docs,
maxlen=max_length,
padding='post')
y_dev = data_loader.get_y_dev()
y_dev_categorical = keras.utils.to_categorical(y_dev, num_classes=2)
print("Fitting model... ")
model.fit(X_train,
y_train,
epochs=3,
batch_size=32,
validation_data=(X_dev, y_dev_categorical),
shuffle=True)
# evaluate the model
y_dev_pred = model.predict(X_dev)
y_dev_pred = np.argmax(y_dev_pred, axis=1)
# loss, accuracy = model.evaluate(X_train, y_train, verbose=0)
# print('Accuracy on training set: ', accuracy)
print('Accuracy on development set: ',
accuracy_score(y_dev, y_dev_pred))
print('Precision on development set: ',
precision_score(y_dev, y_dev_pred))
print('F1-Score on development set: ', f1_score(y_dev, y_dev_pred))
print('Recall on development set: ', recall_score(y_dev, y_dev_pred))
print('Zero percentage on development set: ',
sum([1 for y in y_dev_pred if y == 0]) / len(y_dev_pred))
encoded_test_docs = self.tokenizer.texts_to_sequences(test_docs)
X_test = pad_sequences(encoded_test_docs,
maxlen=max_length,
padding='post')
y_test_pred = model.predict(X_test)
y_test_pred = np.argmax(y_test_pred, axis=1)
# y_test_pred = model.predict_classes(X_test)
print('Zero percentage on test set: ',
sum([1 for y in y_test_pred if y == 0]) / len(y_test_pred))
return y_test_pred
