axis_array[row_arg, col_arg].imshow(image, cmap=cmap)
axis_array[row_arg, col_arg].set_title(titles[image_arg])
image_arg = image_arg + 1
plt.tight_layout()
if __name__ == '__main__':
# from utils.data_manager import DataManager
from utils.utils import get_labels
from keras.models import load_model
import pickle
# dataset_name = 'fer2013'
# model_path = '../trained_models/emotion_models/simple_CNN.985-0.66.hdf5'
dataset_name = 'fer2013'
class_decoder = get_labels(dataset_name)
# data_manager = DataManager(dataset_name)
# faces, emotions = data_manager.get_data()
faces = pickle.load(open('faces.pkl', 'rb'))
emotions = pickle.load(open('emotions.pkl', 'rb'))
pretty_imshow(plt.gca(), make_mosaic(faces[:4], 2, 2), cmap='gray')
plt.show()
"""
image_arg = 0
face = faces[image_arg:image_arg + 1]
emotion = emotions[image_arg:image_arg + 1]
display_image(face, emotion, class_decoder)
plt.show()
normal_imshow(plt.gca(), make_mosaic(faces[:4], 3, 3), cmap='gray')
plt.show()
axis_array[row_arg, col_arg].axis('off')
axis_array[row_arg, col_arg].imshow(image, cmap=cmap)
axis_array[row_arg, col_arg].set_title(titles[image_arg])
image_arg = image_arg + 1
plt.tight_layout()
if __name__ == '__main__':
#from utils.data_manager import DataManager
from utils.utils import get_labels
from keras.models import load_model
import pickle
#dataset_name = 'fer2013'
#model_path = '../trained_models/emotion_models/simple_CNN.985-0.66.hdf5'
dataset_name = 'fer2013'
class_decoder = get_labels(dataset_name)
#data_manager = DataManager(dataset_name)
#faces, emotions = data_manager.get_data()
faces = pickle.load(open('faces.pkl', 'rb'))
emotions = pickle.load(open('emotions.pkl', 'rb'))
pretty_imshow(plt.gca(), make_mosaic(faces[:4], 2, 2), cmap='gray')
plt.show()
"""
image_arg = 0
face = faces[image_arg:image_arg + 1]
emotion = emotions[image_arg:image_arg + 1]
display_image(face, emotion, class_decoder)
plt.show()
normal_imshow(plt.gca(), make_mosaic(faces[:4], 3, 3), cmap='gray')
def main(hparams):
try:
logger.handlers.clear()
except:
pass
logger_name = "SVM - Cross Validation"
logger = logging.getLogger(logger_name)
logger.setLevel("INFO")
file_handler = logging.StreamHandler(sys.stdout)
log_format = '%(asctime)s - %(name)s - %(levelname)s - %(message)s'
formatter = logging.Formatter(log_format)
file_handler.setFormatter(formatter)
logger.addHandler(file_handler)
metrics = {
"eval": {
"acc": [],
"f1": [],
"precision": [],
"recall": [],
"mcc": []
},
"test": {
"acc": [],
"f1": [],
"precision": [],
"recall": [],
"mcc": []
}
}
logger.info("Hyperparameter:")
print(json.dumps(vars(hparams), indent=4))
df_test = pd.read_csv("{}/gold-standard-testset.csv".format(
hparams.data_path))
if hparams.amount_labels == 1:
label_cols = ["GH"]
else:
label_cols = get_labels(hparams.amount_labels)
for seed in range(5):
logger.info(
"Starting Classification for training split {}".format(seed))
logger.info("Loading Text and Labels - use augmentations: {}".format(
hparams.augment))
data_path_split = "{}/{}/".format(hparams.data_path, seed)
if hparams.augment:
df_train = pd.read_csv(
"{}df_train_{}_augmented_{}labels.csv".format(
data_path_split, seed, hparams.amount_labels))
else:
assert hparams.amount_labels == 21
df_train = pd.read_csv("{}df_train_{}.csv".format(
data_path_split, seed))
df_eval = pd.read_csv("{}df_eval_{}.csv".format(data_path_split, seed))
text_train = df_train.text.values
text_eval = df_eval.text.values
text_test = df_test.text.values
labels_train = df_train.loc[:, label_cols].values
labels_eval = df_eval.loc[:, label_cols].values
labels_test = df_test.loc[:, label_cols].values
logger.info("Computing Features")
vectorizer = TfidfVectorizer(max_features=10000, ngram_range=(1, 3))
features_train = vectorizer.fit_transform(text_train)
features_eval = vectorizer.transform(text_eval)
features_test = vectorizer.transform(text_test)
logger.info("Starting Training")
strategy = BinaryRelevance
model = SVC(verbose=True, class_weight="balanced")
classifier = strategy(model)
classifier.fit(features_train, labels_train)
print("\n")
logger.info("Starting Prediction for Evaluation and Testset")
predictions_eval = classifier.predict(features_eval).todense()
predictions_test = classifier.predict(features_test).todense()
logger.info("Metrics on Evaluationset")
metrics_eval = scoring(labels_eval, predictions_eval)
logger.info("Metrics on Testset")
metrics_test = scoring(labels_test, predictions_test)
for metric, value in metrics_eval.items():
metrics["eval"][metric].append(value)
for metric, value in metrics_test.items():
metrics["test"][metric].append(value)
print(
"--------------------------------------------------------------------------------"
)
logger.info("Cross Validation complete")
logger.info("Averaged Metrics on Evaluationset")
metrics_eval = metrics["eval"]
for metric, value in metrics_eval.items():
logger.info("{}: {}".format(metric, np.array(value).mean()))
logger.info("Averaged Metrics on Testset")
metrics_test = metrics["test"]
for metric, value in metrics_test.items():
logger.info("{}: {}".format(metric, np.array(value).mean()))
logger.handlers.clear()
def prepare_data(self):
split_seed = self.hparams.data_path.strip("/").split("/")[-1]
if split_seed != "test":
train_data_path = "{}/df_train_{}_augmented_{}labels.csv".format(
self.hparams.data_path, split_seed, self.hparams.amount_labels)
eval_data_path = "{}/df_eval_{}.csv".format(
self.hparams.data_path, split_seed)
else:
train_data_path = "{}/df_test.csv".format(self.hparams.data_path)
eval_data_path = "{}/df_test.csv".format(self.hparams.data_path)
train_eval_path = "/".join(
self.hparams.data_path.strip("/").split("/")[:-1])
test_data_path = "{}/gold-standard-testset.csv".format(train_eval_path)
training_set = pd.read_csv(train_data_path)
evaluation_set = pd.read_csv(eval_data_path)
try:
test_set = pd.read_csv(test_data_path)
except:
test_set = pd.read_csv("/" + test_data_path)
labels = get_labels(self.hparams.amount_labels)
max_length = 4096
print("Computing Input")
training_inputs = [
self.tokenizer(text,
max_length=max_length,
padding="max_length",
truncation=True)
for text in tqdm(training_set.text.values, total=len(training_set))
]
training_input_ids = [
training_input["input_ids"] for training_input in training_inputs
]
training_attention_mask = [
training_input["attention_mask"]
for training_input in training_inputs
]
training_labels = training_set.loc[:, labels].values
evaluation_inputs = [
self.tokenizer(text,
max_length=max_length,
padding="max_length",
truncation=True)
for text in tqdm(evaluation_set.text.values,
total=len(evaluation_set))
]
evaluation_input_ids = [
evaluation_input["input_ids"]
for evaluation_input in evaluation_inputs
]
evaluation_attention_mask = [
evaluation_input["attention_mask"]
for evaluation_input in evaluation_inputs
]
evaluation_labels = evaluation_set.loc[:, labels].values
test_inputs = [
self.tokenizer(text,
max_length=max_length,
padding="max_length",
truncation=True)
for text in tqdm(test_set.text.values, total=len(test_set))
]
test_input_ids = [
test_input["input_ids"] for test_input in test_inputs
]
test_attention_mask = [
test_input["attention_mask"] for test_input in test_inputs
]
test_labels = test_set.loc[:, labels].values
training_input_ids = torch.tensor(training_input_ids)
training_attention_mask = torch.tensor(training_attention_mask)
training_labels = torch.tensor(training_labels)
evaluation_input_ids = torch.tensor(evaluation_input_ids)
evaluation_attention_mask = torch.tensor(evaluation_attention_mask)
evaluation_labels = torch.tensor(evaluation_labels)
test_input_ids = torch.tensor(test_input_ids)
test_attention_mask = torch.tensor(test_attention_mask)
test_labels = torch.tensor(test_labels)
self.label_weights = 1 / \
evaluation_labels.shape[0] * evaluation_labels.sum(dim=0)
self.train_data = TensorDataset(training_input_ids,
training_attention_mask,
training_labels)
self.valid_data = TensorDataset(evaluation_input_ids,
evaluation_attention_mask,
evaluation_labels)
self.test_data = TensorDataset(test_input_ids, test_attention_mask,
test_labels)
def prepare_data(self):
split_seed = self.hparams.data_path.strip("/").split("/")[-1]
if split_seed != "test":
if self.hparams.label is not None:
train_data_path = "{}/df_train_{}{}_augmented_{}labels.csv".format(
self.hparams.data_path, self.hparams.label.lower(),
split_seed, self.hparams.amount_labels)
else:
train_data_path = "{}/df_train_{}_augmented_{}labels.csv".format(
self.hparams.data_path, split_seed,
self.hparams.amount_labels)
eval_data_path = "{}/df_eval_{}.csv".format(
self.hparams.data_path, split_seed)
else:
train_data_path = "{}/df_test.csv".format(self.hparams.data_path)
eval_data_path = "{}/df_test.csv".format(self.hparams.data_path)
train_eval_path = "/".join(
self.hparams.data_path.strip("/").split("/")[:-1])
test_data_path = "{}/gold-standard-testset.csv".format(train_eval_path)
try:
training_set = pd.read_csv(train_data_path)
except:
if self.hparams.augment:
print("Augmenting Data...")
train_data_path_unaug = "{}/df_train_{}.csv".format(
self.hparams.data_path, split_seed)
unaug_training_set = pd.read_csv(train_data_path_unaug)
augmentations_path = "/".join(
self.hparams.data_path.strip("/").split("/")[:-2])
augmentations_df = pd.read_csv(
"/{}/article_confirmed_summary_augmentations.csv".format(
augmentations_path))
training_set = augment_dataframe(
df=unaug_training_set,
augmentations_df=augmentations_df,
categories=[self.hparams.label.upper()])
training_set.to_csv(train_data_path, index=False)
else:
print("Loading Data without Augmentation...")
train_data_path = "{}/df_train_{}.csv".format(
self.hparams.data_path, split_seed)
training_set = pd.read_csv(train_data_path)
training_set.loc[:, "none"] = 0
for index, row in training_set.iterrows():
if row.loc["none"] == 0:
training_set.loc[index, "none"] = 1
evaluation_set = pd.read_csv(eval_data_path)
try:
test_set = pd.read_csv(test_data_path)
except:
test_set = pd.read_csv("/" + test_data_path)
if self.hparams.label is not None:
not_label = "Not{}".format(self.hparams.label.upper())
training_set.rename(columns={"none": not_label}, inplace=True)
evaluation_set.loc[:, not_label] = 0
for index, row in evaluation_set.iterrows():
if row.UD == 0:
evaluation_set.loc[index, not_label] = 1
test_set.loc[:, not_label] = 0
for index, row in test_set.iterrows():
if row.UD == 0:
test_set.loc[index, not_label] = 1
labels = [self.hparams.label.upper(), not_label]
else:
labels = get_labels(self.hparams.amount_labels)
max_length = 512
data_column = "summary_{}".format(self.hparams.summary_type)
print("Computing Input")
training_inputs = [
self.tokenizer(text,
max_length=max_length,
padding="max_length",
truncation=True)
for text in tqdm(training_set.loc[:, data_column].values,
total=len(training_set))
]
training_input_ids = [
training_input["input_ids"] for training_input in training_inputs
]
training_attention_mask = [
training_input["attention_mask"]
for training_input in training_inputs
]
training_labels = training_set.loc[:, labels].values
evaluation_inputs = [
self.tokenizer(text,
max_length=max_length,
padding="max_length",
truncation=True)
for text in tqdm(evaluation_set.loc[:, data_column].values,
total=len(evaluation_set))
]
evaluation_input_ids = [
evaluation_input["input_ids"]
for evaluation_input in evaluation_inputs
]
evaluation_attention_mask = [
evaluation_input["attention_mask"]
for evaluation_input in evaluation_inputs
]
evaluation_labels = evaluation_set.loc[:, labels].values
test_inputs = [
self.tokenizer(text,
max_length=max_length,
padding="max_length",
truncation=True)
for text in tqdm(test_set.loc[:, data_column].values,
total=len(test_set))
]
test_input_ids = [
test_input["input_ids"] for test_input in test_inputs
]
test_attention_mask = [
test_input["attention_mask"] for test_input in test_inputs
]
test_labels = test_set.loc[:, labels].values
training_input_ids = torch.tensor(training_input_ids)
training_attention_mask = torch.tensor(training_attention_mask)
training_labels = torch.tensor(training_labels)
evaluation_input_ids = torch.tensor(evaluation_input_ids)
evaluation_attention_mask = torch.tensor(evaluation_attention_mask)
evaluation_labels = torch.tensor(evaluation_labels)
test_input_ids = torch.tensor(test_input_ids)
test_attention_mask = torch.tensor(test_attention_mask)
test_labels = torch.tensor(test_labels)
self.label_weights = 1 / \
evaluation_labels.shape[0] * evaluation_labels.sum(dim=0)
self.train_data = TensorDataset(training_input_ids,
training_attention_mask,
training_labels)
self.valid_data = TensorDataset(evaluation_input_ids,
evaluation_attention_mask,
evaluation_labels)
self.test_data = TensorDataset(test_input_ids, test_attention_mask,
test_labels)
def prepare_data(self):
split_seed = self.hparams.data_path.strip("/").split("/")[-1]
if split_seed != "test":
train_data_path = "{}/df_train_{}_augmented_{}labels.csv".format(
self.hparams.data_path, split_seed, self.hparams.amount_labels)
eval_data_path = "{}/df_eval_{}.csv".format(
self.hparams.data_path, split_seed)
else:
train_data_path = "{}/df_test.csv".format(self.hparams.data_path)
eval_data_path = "{}/df_test.csv".format(self.hparams.data_path)
train_eval_path = "/".join(
self.hparams.data_path.strip("/").split("/")[:-1])
test_data_path = "{}/gold-standard-testset.csv".format(train_eval_path)
training_set = pd.read_csv(train_data_path)
evaluation_set = pd.read_csv(eval_data_path)
try:
test_set = pd.read_csv(test_data_path)
except:
test_set = pd.read_csv("/" + test_data_path)
labels = get_labels(self.hparams.amount_labels)
split_length = self.hparams.split_size
shift = self.hparams.shift
self.tokenizer.padding_side = "right"
self.tokenizer.pad_token = "<pad>"
with IncreasedLoggingLevel("transformers.tokenization_utils_base"):
max_length_training = max([
len(self.tokenizer(text)["input_ids"])
for text in training_set.text.values
])
max_length_evaluation = max([
len(self.tokenizer(text)["input_ids"])
for text in evaluation_set.text.values
])
max_length_test = max([
len(self.tokenizer(text)["input_ids"])
for text in test_set.text.values
])
max_padding_training = ceil(max_length_training/split_length) * \
split_length + split_length
max_padding_evaluation = ceil(max_length_evaluation/split_length) * \
split_length + split_length
max_padding_test = ceil(max_length_test/split_length) * \
split_length + split_length
print("Computing Input")
with IncreasedLoggingLevel("transformers.tokenization_utils_base"):
training_inputs = [
self.tokenizer(text,
max_length=max_padding_training,
padding="max_length",
truncation=True)
for text in tqdm(training_set.text.values,
total=len(training_set))
]
training_input_ids, training_attention_mask = split_articles(
training_inputs,
max_length=max_padding_training,
split_length=split_length,
shift=shift)
training_labels = training_set.loc[:, labels].values
with IncreasedLoggingLevel("transformers.tokenization_utils_base"):
evaluation_inputs = [
self.tokenizer(text,
max_length=max_padding_evaluation,
padding="max_length",
truncation=True)
for text in tqdm(evaluation_set.text.values,
total=len(evaluation_set))
]
evaluation_input_ids, evaluation_attention_mask = split_articles(
evaluation_inputs,
max_length=max_padding_evaluation,
split_length=split_length,
shift=shift)
evaluation_labels = evaluation_set.loc[:, labels].values
with IncreasedLoggingLevel("transformers.tokenization_utils_base"):
test_inputs = [
self.tokenizer(text,
max_length=max_padding_test,
padding="max_length",
truncation=True)
for text in tqdm(test_set.text.values, total=len(test_set))
]
test_input_ids, test_attention_mask = split_articles(
test_inputs,
max_length=max_padding_test,
split_length=split_length,
shift=shift)
test_labels = test_set.loc[:, labels].values
training_input_ids = torch.tensor(training_input_ids)
training_attention_mask = torch.tensor(training_attention_mask)
training_labels = torch.tensor(training_labels)
evaluation_input_ids = torch.tensor(evaluation_input_ids)
evaluation_attention_mask = torch.tensor(evaluation_attention_mask)
evaluation_labels = torch.tensor(evaluation_labels)
test_input_ids = torch.tensor(test_input_ids)
test_attention_mask = torch.tensor(test_attention_mask)
test_labels = torch.tensor(test_labels)
self.label_weights = 1 / \
evaluation_labels.shape[0] * evaluation_labels.sum(dim=0)
self.train_data = TensorDataset(training_input_ids,
training_attention_mask,
training_labels)
self.valid_data = TensorDataset(evaluation_input_ids,
evaluation_attention_mask,
evaluation_labels)
self.test_data = TensorDataset(test_input_ids, test_attention_mask,
test_labels)
