default="https://www.google.com",
help="Website to test")
args = parser.parse_args()
tokenizerFolder = args.tokenizer_folder
savedModelDirectory = args.model_dir
websiteToTest = args.website_to_test
threshold = args.threshold
# Loading files
# Load tokenization files
tokenizer = ByteLevelBPETokenizer(
tokenizerFolder + "/tokenizer.tok-vocab.json",
tokenizerFolder + "/tokenizer.tok-merges.txt",
)
tokenizerVocabSize = tokenizer.get_vocab_size()
print("Tokenizer files have been loaded and the vocab size is %d..." %
tokenizerVocabSize)
# Load saved model
model = load(savedModelDirectory + "/phishytics-model.joblib")
print("Model loaded...")
# Load document frequency dictionary
docDict = np.load(savedModelDirectory +
"/phishytics-model-tfidf-dictionary.npy",
allow_pickle=True).item()
print("Document frequency dictionary loaded...")
# Testing
print("Loading webpage...")
def architecture_search(process_id):
os.makedirs(f"checkpoints/{process_id+1}")
os.makedirs(f"tokenizer/{process_id+1}")
files = glob.glob("../../data/pre_abstract_txts/*.txt")
tok_sizes = list(range(100, 2000, 100))
hidden_sizes = list(range(12, 300, 12))
emb_sizes = list(range(10, 250, 10))
cased = [True, False]
batch_size = 1
results = {}
choices = list(itertools.product(tok_sizes, hidden_sizes, emb_sizes,
cased))
random.shuffle(choices)
best_acc = -np.inf
while len(choices) > 0:
tok_size, hidden_size, emb_size, cased = choices.pop()
print(tok_size, hidden_size, emb_size, cased)
tokenizer = ByteLevelBPETokenizer(lowercase=cased)
tokenizer.train(files, vocab_size=tok_size, special_tokens=["[PAD]"])
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
dataset = TextDataset(data_dir="../../data/pre_abstract_txts",
labels_dir="../../data/pre_abstract_labels",
device=device,
tokenizer=tokenizer,
batch_size=batch_size)
test_dataset = TextDataset(
data_dir="../../data/pre_abstract_txts",
labels_dir="../../data/pre_abstract_labels_test",
device=device,
tokenizer=tokenizer,
batch_size=batch_size)
model = LSTMTagger(vocab_size=tokenizer.get_vocab_size(),
embedding_dim=emb_size,
lstm_dim=hidden_size,
dropout=0,
n_classes=len(dataset.classes)).to(device)
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters())
# optimizer = torch.optim.SGD(model.parameters(), momentum=0.9, nesterov=True, lr=v)
epoch = 0
n = 3
test_acc = -np.inf
log_interval = 10 # all n batches
weights = copy.deepcopy(model.state_dict())
while True:
dataset.shuffle()
epoch += 1
model.train()
total_loss = 0.
pbar = tqdm.tqdm(enumerate(dataset), desc=f"epoch {epoch}")
for i, (x, y) in pbar:
# reset gradients
optimizer.zero_grad()
# feed forward batch
output = model(x)
# calculate loss
loss = criterion(output.transpose(1, 2), y)
# back propagate loss
loss.backward()
# norm and clip gradients
# torch.nn.utils.clip_grad_norm_(model.parameters(), 0.5)
optimizer.step()
pbar.set_description(
f'epoch {epoch} | batch {i + 1:d}/{len(dataset)} | loss {loss.item():.2f}'
)
model.eval()
a, c = 0, 0
with torch.no_grad():
t_loss = 0
for i, (x, y) in enumerate(test_dataset):
output = model(x)
loss = criterion(output.transpose(1, 2), y)
t_loss += loss.item()
for p, t in zip(torch.argmax(output, -1), y):
for pi, ti in zip(p, t):
a += 1
if pi == ti:
c += 1
acc = c / a
if acc <= test_acc and n > 0:
n -= 1
continue
elif acc <= test_acc:
break
print(t_loss, acc)
weights = copy.deepcopy(model.state_dict())
test_acc = acc
results[(tok_size, hidden_size, emb_size, cased)] = acc
print(
list(
sorted([(k, v) for k, v in results.items()],
key=lambda y: y[1],
reverse=True))[:10])
print(best_acc, test_acc)
if test_acc > best_acc:
best_acc = test_acc
dir_path = f"tokenizer/{process_id+1}/lstm-tagger-{best_acc:.6f}"
if os.path.exists(dir_path):
continue
torch.save(
weights,
f"checkpoints/{process_id+1}/lstm-tagger-{best_acc:.6f}.pt")
os.makedirs(dir_path)
tokenizer.save(dir_path)
class HuggingFaceBpeHelper(BPEHelper):
"""
HuggingFace's ByteLevelBPE Tokenizer.
Fast because Rust.
"""
def __init__(self, opt: Opt, shared: TShared = None):
super().__init__(opt, shared)
# Default true for HF
self.special_tok_map = {} # map from HF
self.add_prefix_space = opt.get('bpe_add_prefix_space', True)
if self.add_prefix_space is None:
self.add_prefix_space = True
if opt.get('dict_loaded'):
dfname = opt['dict_file']
if PathManager.exists(f'{dfname}-merges.txt'):
opt['bpe_merge'] = f'{dfname}-merges.txt'
if PathManager.exists(f'{dfname}-vocab.json'):
opt['bpe_vocab'] = f'{dfname}-vocab.json'
try:
from tokenizers import ByteLevelBPETokenizer
except ImportError:
raise ImportError(
'Please install HuggingFace tokenizer with: pip install tokenizers'
)
if self.bpe_dropout:
raise NotImplementedError(
'--bpe-dropout is not supported with ByteLevelBPE because tokenizers '
'library does not allow dynamically turning BPE on/off. You can use '
'--dict-tokenizer slow_bytelevel_bpe to gain this feature.'
)
if self.lower:
warn_once('Are you sure you want to lower case your BPE dictionary?')
if self.maxtokens > 0 or self.minfreq > 0:
raise ValueError(
'You should not filter vocabulary with using --dict-tokenizer bytelevelbpe'
' (no --dict-minfreq or --dict-maxtokens).'
)
if 'bpe_vocab' not in opt:
raise ValueError('--bpe-vocab is required for loading pretrained tokenizer')
if 'bpe_merge' not in opt:
raise ValueError('--bpe-merge is required for loading pretrained tokenizer')
self.vocab_path = opt['bpe_vocab']
self.merge_path = opt['bpe_merge']
if not self.vocab_path or not self.merge_path:
raise IOError(
'--bpe-vocab and --bpe-merge are mandatory with '
'--dict-tokenizer bytelevelbpe'
)
if not PathManager.exists(self.vocab_path):
raise IOError(
f'File {self.vocab_path} does not exist. --bpe-vocab must be pretrained.'
)
if not PathManager.exists(self.merge_path):
raise IOError(
f'File {self.merge_path} does not exist. --bpe-merge must be pretrained.'
)
self.tokenizer = ByteLevelBPETokenizer(
self.vocab_path, self.merge_path, self.add_prefix_space
)
def helper_encode(self, text: str) -> List[str]:
"""
Decode list of tokens into text string.
:param tokens:
list of tokens
:param delimiter:
string delimiter for tokens
:return text:
decoded text
"""
return self.tokenizer.encode(text).tokens
def helper_decode(
self, tokens: List[str], token_ids: List[int], delimiter: str
) -> str:
"""
Decode list of tokens into text string.
:param tokens:
list of tokens
:param token_ids:
list of token ids
:param delimiter:
string delimiter for tokens
:return text:
decoded text
"""
text = self.tokenizer.decode(token_ids, skip_special_tokens=False)
return text
def add_special_tokens(self, dict_agent, special_tokens: List[str]):
"""
Add special tokens to the tokenizer and dict_agent.
"""
logging.debug(f'adding the following special tokens: {special_tokens}')
self.tokenizer.add_special_tokens(special_tokens) # add to HF
for tok in special_tokens:
parlai_key = dict_agent[tok]
hf_key = self.tokenizer.token_to_id(tok)
self.special_tok_map[parlai_key] = hf_key
def sync_with_dict(self, dict_agent):
"""
Sync the dictionary agent with Hugging Face tokenizer's BPE dict.
Called only once on initialization.
"""
special_tokens = [
dict_agent.null_token,
dict_agent.start_token,
dict_agent.end_token,
dict_agent.unk_token,
]
self.add_special_tokens(dict_agent, special_tokens)
for i in range(self.tokenizer.get_vocab_size() - len(special_tokens)):
token = self.tokenizer.id_to_token(i)
dict_agent.add_token(token)
# We don't have access to the hugging face word frequency table,
# just set it to 1 instead
dict_agent.freq[token] = 1
def save(self, dir_name: str, file_name: str):
"""
Save appropriate files.
:param dir_name:
directory to save.
:param file_name:
file to save.
"""
self.tokenizer.save_model(dir_name, file_name)
for (_, _, f) in walk(labeledDataFolder + "/legitimate_htmls"):
files.extend(
[labeledDataFolder + "/legitimate_htmls/" + file for file in f])
for (_, _, f) in walk(labeledDataFolder + "/phishing_htmls"):
files.extend([labeledDataFolder + "/phishing_htmls/" + file for file in f])
print("Total number of html files: %d\n" % len(files))
# Writing data, one html file per line. This is the format the tokenizer expects
print("Writing html data into a single file...")
output = open("tokenizer/htmlCodePerLine.txt", "w")
count = 0
for file in files:
count = count + 1
print("Files processed: %d, Total files: %d" % (count, len(files)))
fileData = io.open(file, "r", errors="ignore").readlines()
fileData = ''.join(str(line) for line in fileData)
fileData = fileData.replace("\n", " ")
output.write(fileData + "\n")
output.close()
# Starting tokenization
print("\nStarting tokenization with BPE")
tokenizer = ByteLevelBPETokenizer()
tokenizer.train("tokenizer/htmlCodePerLine.txt",
min_frequency=minFrequency,
vocab_size=vocabSize)
print(
"Vocabulary size is: %d\nNOTE: Sometimes, the vocab size might not be equal to the input 'vocab_size'\n"
% (tokenizer.get_vocab_size()))
tokenizer.save("tokenizer", "tokenizer.tok")
print("Tokenizer files have been saved in 'tokenizer' directory...")
class HuggingFaceBpeHelper(BPEHelper):
"""
HuggingFace's ByteLevelBPE Tokenizer.
Fast because Rust.
"""
def __init__(self, opt: Opt, shared: TShared = None):
super().__init__(opt, shared)
# Default true for HF
self.add_prefix_space = opt.get('bpe_add_prefix_space', True)
if self.add_prefix_space is None:
self.add_prefix_space = True
if opt.get('dict_loaded'):
dfname = opt['dict_file']
if os.path.isfile(f'{dfname}-merges.txt'):
opt['bpe_merge'] = f'{dfname}-merges.txt'
if os.path.isfile(f'{dfname}-vocab.json'):
opt['bpe_vocab'] = f'{dfname}-vocab.json'
try:
from tokenizers import ByteLevelBPETokenizer
except ImportError:
raise ImportError(
'Please install HuggingFace tokenizer with: pip install tokenizers'
)
if self.lower:
raise ValueError(
'Only use --dict-lower false with --dict-tokenizer bytelevelbpe'
)
if self.maxtokens > 0 or self.minfreq > 0:
raise ValueError(
'You should not filter vocabulary with using --dict-tokenizer bytelevelbpe'
' (no --dict-minfreq or --dict-maxtokens).')
if 'bpe_vocab' not in opt:
raise ValueError(
'--bpe-vocab is required for loading pretrained tokenizer')
if 'bpe_merge' not in opt:
raise ValueError(
'--bpe-merge is required for loading pretrained tokenizer')
self.vocab_path = opt['bpe_vocab']
self.merge_path = opt['bpe_merge']
if not self.vocab_path or not self.merge_path:
raise IOError('--bpe-vocab and --bpe-merge are mandatory with '
'--dict-tokenizer bytelevelbpe')
if not os.path.isfile(self.vocab_path):
raise IOError(
f'File {self.vocab_path} does not exist. --bpe-vocab must be pretrained.'
)
if not os.path.isfile(self.merge_path):
raise IOError(
f'File {self.merge_path} does not exist. --bpe-merge must be pretrained.'
)
self.tokenizer = ByteLevelBPETokenizer(self.vocab_path,
self.merge_path,
self.add_prefix_space)
def helper_encode(self, text: str) -> List[str]:
"""
Decode list of tokens into text string.
:param tokens:
list of tokens
:param delimiter:
string delimiter for tokens
:return text:
decoded text
"""
return self.tokenizer.encode(text).tokens
def helper_decode(self, tokens: List[str], token_ids: List[int],
delimiter: str) -> str:
"""
Decode list of tokens into text string.
:param tokens:
list of tokens
:param token_ids:
list of token ids
:param delimiter:
string delimiter for tokens
:return text:
decoded text
"""
text = self.tokenizer.decode(token_ids)
return text
def sync_with_dict(self, dict_agent):
"""
Sync the dictionary agent with Hugging Face tokenizer's BPE dict.
Called only once on initialization.
"""
special_tokens = [
dict_agent.null_token,
dict_agent.start_token,
dict_agent.end_token,
dict_agent.unk_token,
]
self.tokenizer.add_special_tokens(special_tokens)
for i in range(self.tokenizer.get_vocab_size() - 4):
token = self.tokenizer.id_to_token(i)
dict_agent.add_token(token)
# We don't have access to the hugging face word frequency table,
# just set it to 1 instead
dict_agent.freq[token] = 1
def save(self, dir_name: str, file_name: str):
"""
Save appropriate files.
:param dir_name:
directory to save.
:param file_name:
file to save.
"""
self.tokenizer.save(dir_name, file_name)
class CodeTrainedBPE_Translation_DataProcessor(DataProcessor, Dataset):
def __init__(self, task_data, max_src_len=512, max_tgt_len=512):
"""
This data processor tokenizes and numericalises using a custom byte pair
encoding trained on the codeSearchNet train data with full docstrings.
"""
self.task_data = task_data
self.max_src_len = max_src_len
self.max_tgt_len = max_tgt_len
self.tokenizer = ByteLevelBPETokenizer(
"/nfs/phd_by_carlos/notebooks/datasets/code_search_net/code_bpe_hugging_32k-vocab.json",
"/nfs/phd_by_carlos/notebooks/datasets/code_search_net/code_bpe_hugging_32k-merges.txt"
)
self.tokenizer.add_special_tokens(["[CLS]", "[SOS]", "[EOS]", "[PAD]"])
self.SOS = self.tokenizer.encode("[SOS]").ids[0]
self.EOS = self.tokenizer.encode("[EOS]").ids[0]
self.PAD = self.tokenizer.encode("[PAD]").ids[0]
self.CLS = self.tokenizer.encode("[CLS]").ids[0]
self.__remove_long_samples()
def __len__(self):
return len(self.task_data)
def __getitem__(self, idx):
src, tgt = self.task_data[idx]
sample = {'src': self.encode(src), 'tgt': self.encode(tgt)}
return sample
@property
def vocab_size(self):
return self.tokenizer.get_vocab_size()
def __remove_long_samples(self):
for i in tqdm.tqdm(list(reversed(range(len(self.task_data)))),
desc="removing long samples"):
src, tgt = self.task_data[i]
if len(self.encode(src)) > self.max_src_len or len(
self.encode(tgt)) > self.max_tgt_len:
del self.task_data[i]
def encode(self, sample):
"""
sample: str: the input string to encode
"""
return [self.SOS] + self.tokenizer.encode(sample).ids + [self.EOS]
def encode_src(self, sample):
return self.encode(sample)
def encode_tgt(self, sample):
return self.encode(sample)
def encode_to_tensor(self, input_samples):
"""
input_samples: [str]: one or more strings to convert to a single padded tensor. (Seq_len x batch)
"""
return pad_sequence([
torch.Tensor(self.encode(sample)).type(torch.LongTensor)
for sample in input_samples
],
padding_value=self.PAD)
def collate(self, input_samples):
"""
input_samples: [dict]: these are samples obtained through the _get_item method
"""
collated_samples = {}
sample_keys = input_samples[0].keys()
for key in sample_keys:
collated_samples[key] = torch.nn.utils.rnn.pad_sequence(
[
torch.Tensor(sample[key]).type(torch.LongTensor)
for sample in input_samples
],
padding_value=self.PAD)
return collated_samples
def decode(self, ids):
"""
ids: [int]: ids to decode
"""
return self.tokenizer.decode(ids)
def decode_src(self, ids):
return self.decode(ids)
def decode_tgt(self, ids):
return self.decode(ids)
def validate_prediction(self, numerical_sequence):
# there are no constraints
return True
def prediction_is_complete(self, numerical_sequence):
return self.EOS in numerical_sequence
def decode_tensor(self, output_tensor):
"""
output_tensor: [[int]]: model output (Seq_len x batch)
"""
batch_first_output_tensor = output_tensor.T
return [
self.decode(sequence.cpu().tolist())
for sequence in batch_first_output_tensor
]
def to_dataloader(self,
batch_size,
repeat=False,
num_workers=4,
shuffle=True):
"""
This function returns an iterable object with all the data batched.
>>> BPE_processor = CodeTrainedBPE_Translation_DataProcessor(validation_pairs, max_tgt_len=100)
>>> dataloader = BPE_processor.to_dataloader(2)
>>> for i_batch, sample_batched in enumerate(dataloader):
>>> print(sample_batched["tgt"])
>>> print(BPE_processor.decode_tensor(sample_batched["tgt"]))
>>> break
"""
return DataLoader(self, batch_size=batch_size, num_workers=num_workers,\
drop_last=False, collate_fn = self.collate, shuffle=shuffle)
def save(self, path):
torch.save(self, path)
class Parse_Tree_Translation_DataProcessor(Dataset):
def __init__(
self,
task_data,
max_length=500,
tokenizer_dir="/nfs/phd_by_carlos/notebooks/datasets/code_search_net/",
grammar_path="src/tree-sitter/tree-sitter-python/src/grammar.json",
**kwargs):
self.task_data = task_data
self.max_length = max_length
self.tokenizer = ByteLevelBPETokenizer(
tokenizer_dir + "code_bpe_hugging_32k-vocab.json",
tokenizer_dir + "code_bpe_hugging_32k-merges.txt")
self.tokenizer.add_special_tokens(["[CLS]", "[SOS]", "[EOS]", "[PAD]"])
self.SOS = self.tokenizer.encode("[SOS]").ids[0]
self.EOS = self.tokenizer.encode("[EOS]").ids[0]
self.PAD = self.tokenizer.encode("[PAD]").ids[0]
self.CLS = self.tokenizer.encode("[CLS]").ids[0]
with open(grammar_path, "r") as grammar_file:
self.python_grammar = json.load(grammar_file)
extra_externals = {
"_string_start": {
"type": "PATTERN",
"value": '"'
},
"_string_content": {
"type": "PATTERN",
"value": "[A-Za-z0-9 _,.()\/{}!$@'*]*"
},
"_string_end": {
"type": "PATTERN",
"value": '"'
},
"_newline": {
"type": "BLANK"
}
}
for node_type, member in extra_externals.items():
self.python_grammar["rules"][node_type] = member
self.python_parser = Code_Parser(self.python_grammar, "python",
**kwargs)
self.node_processor = Node_Processor()
self.tree_vocab, grammar_patterns = get_grammar_vocab(
self.python_grammar)
self.tokenizer.add_tokens(["<REDUCE>"])
for tree_token in sorted(self.tree_vocab):
if len(self.tokenizer.encode(tree_token).tokens) != 1:
self.tokenizer.add_tokens([tree_token])
# filtering the data
filtered_task_data = []
for desc, code in self.task_data:
numerical_code_sequence = self.encode_tgt(code)
numerical_desc_sequence = self.encode_src(desc)
token_sequence = self.numerical_to_token_sequence(
numerical_code_sequence)
if self.python_parser.is_valid_sequence(token_sequence) and len(
token_sequence) <= max_length and len(
numerical_desc_sequence) <= max_length:
filtered_task_data.append((desc, code))
elif len(token_sequence) > max_length or len(
numerical_desc_sequence) > max_length:
print(
f"Sequence too long: src->{len(numerical_desc_sequence)}, tgt->{len(token_sequence)}"
)
else:
print(f"Could not parse and reconstruct: {code}")
self.task_data = filtered_task_data
def __len__(self):
return len(self.task_data)
def __getitem__(self, idx):
if idx >= len(self):
raise IndexError
src, tgt = self.task_data[idx]
sample = {'src': self.encode_src(src), 'tgt': self.encode_tgt(tgt)}
return sample
@property
def vocab_size(self):
return self.tokenizer.get_vocab_size()
def encode_src(self, desc_str):
return [self.SOS] + self.tokenizer.encode(desc_str).ids + [self.EOS]
def encode_tgt(self, code_str):
code_sequence = self.python_parser.code_to_sequence(code_str)
numerical_code = []
for code_token in code_sequence:
numerical_code += self.tokenizer.encode(code_token).ids
return [self.SOS] + numerical_code + [self.EOS]
def decode_src(self, numerical_desc):
"""
ids: [int]: ids to decode
"""
return self.tokenizer.decode(ids)
def numerical_to_token_sequence(self, numerical_code):
token_sequence = [
self.tokenizer.decode([token_idx]) for token_idx in numerical_code
if token_idx not in [self.SOS, self.EOS, self.PAD, self.CLS]
]
return token_sequence
def decode_tgt(self, numerical_code):
token_sequence = self.numerical_to_token_sequence(numerical_code)
partial_tree = self.python_parser.sequence_to_partial_tree(
token_sequence)
return self.node_processor.pretty_print(
partial_tree.root), partial_tree
def validate_prediction(self, current_prediction):
# print(f"validating: {current_prediction}")
token_sequence = self.numerical_to_token_sequence(current_prediction)
return self.python_parser.is_valid_sequence(token_sequence)
def prediction_is_complete(self, current_prediction):
token_sequence = self.numerical_to_token_sequence(current_prediction)
return self.python_parser.sequence_to_partial_tree(
token_sequence).is_complete
def collate(self, input_samples):
"""
input_samples: [dict]: these are samples obtained through the _get_item method
"""
collated_samples = {}
sample_keys = input_samples[0].keys()
for key in sample_keys:
collated_samples[key] = torch.nn.utils.rnn.pad_sequence(
[
torch.Tensor(sample[key]).type(torch.LongTensor)
for sample in input_samples
],
padding_value=self.PAD)
return collated_samples
def to_dataloader(self, batch_size, num_workers=4, shuffle=True):
"""
This function returns an iterable object with all the data batched.
>>> BPE_processor = CodeTrainedBPE_Translation_DataProcessor(validation_pairs, max_tgt_len=100)
>>> dataloader = BPE_processor.to_dataloader(2)
>>> for i_batch, sample_batched in enumerate(dataloader):
>>> print(sample_batched["tgt"])
>>> print(BPE_processor.decode_tensor(sample_batched["tgt"]))
>>> break
"""
return DataLoader(self, batch_size=batch_size, num_workers=num_workers,\
drop_last=False, collate_fn = self.collate, shuffle=shuffle)
def save(self, path):
torch.save(self, path)
def __init__(self, path, vocab_size=-1, use_bpe=False, tokenizer_data=""):
self.dictionary = Dictionary()
if use_bpe:
assert os.path.exists(path), "Path does not exist: " + path
print(
"-------------------------------------------------------------"
)
tokenizer = ByteLevelBPETokenizer()
if len(tokenizer_data) != 0:
print("Training tokenizer on: " +
os.path.join(tokenizer_data, 'train.txt'))
tokenizer.train([os.path.join(tokenizer_data, 'train.txt')],
vocab_size=vocab_size,
show_progress=False)
else:
print("Training tokenizer on: " +
os.path.join(path, 'train.txt'))
tokenizer.train(
[
os.path.join(path, 'train.txt')
# os.path.join(path, 'valid.txt'),
# os.path.join(path, 'test.txt')
],
vocab_size=vocab_size,
show_progress=False)
print(
"-------------------------------------------------------------"
)
print("Encoding dataset at: " + path)
with open(os.path.join(path, 'train.txt'), 'r',
encoding='utf-8') as f:
text = f.read()
enc = tokenizer.encode(text)
tokens = len(enc.ids)
ids = torch.LongTensor(tokens)
for index, id in enumerate(enc.ids):
ids[index] = id
self.train = ids
self.dictionary.avg_characters_per_token['train'] = len(
text) / len(enc.ids)
with open(os.path.join(path, 'valid.txt'), 'r',
encoding='utf-8') as f:
text = f.read()
enc = tokenizer.encode(text)
tokens = len(enc.ids)
ids = torch.LongTensor(tokens)
for index, id in enumerate(enc.ids):
ids[index] = id
self.valid = ids
self.dictionary.avg_characters_per_token['valid'] = len(
text) / len(enc.ids)
with open(os.path.join(path, 'test.txt'), 'r',
encoding='utf-8') as f:
text = f.read()
enc = tokenizer.encode(text)
tokens = len(enc.ids)
ids = torch.LongTensor(tokens)
for index, id in enumerate(enc.ids):
ids[index] = id
self.test = ids
self.dictionary.avg_characters_per_token['test'] = len(
text) / len(enc.ids)
print(
"-------------------------------------------------------------"
)
self.dictionary.word2idx = tokenizer.get_vocab()
self.dictionary.idx2word = [
tokenizer.id_to_token(x)
for x in range(tokenizer.get_vocab_size())
]
self.dictionary.total = tokenizer.get_vocab_size()
else:
self.train = self.tokenize(os.path.join(path, 'train.txt'))
self.valid = self.tokenize(os.path.join(path, 'valid.txt'))
self.test = self.tokenize(os.path.join(path, 'test.txt'))
def getURL():
if request.method == 'POST':
urlname = request.form['url']
url = request.form['url']
print(url)
tokenizerFolder = "tokenizer"
savedModelDirectory = "saved_models"
websiteToTest = url
threshold = 0.5
tokenizer = ByteLevelBPETokenizer(
tokenizerFolder + "/tokenizer.tok-vocab.json",
tokenizerFolder + "/tokenizer.tok-merges.txt",
)
tokenizerVocabSize = tokenizer.get_vocab_size()
print("Tokenizer files have been loaded and the vocab size is %d..." % tokenizerVocabSize)
model = load(savedModelDirectory + "/phishytics-model.joblib")
print("Model loaded...")
# Load document frequency dictionary
docDict = np.load(savedModelDirectory + "/phishytics-model-tfidf-dictionary.npy", allow_pickle=True).item()
print("Document frequency dictionary loaded...")
# Testing
print("Loading webpage...")
try:
request1 = requests.get(websiteToTest)
webpageHtml = str(request1.text)
webpageHtml = webpageHtml.replace("\n", " ")
except Exception as e:
print('\n',e)
print("\nAn error occurred, exiting now... ")
exit()
# Convert text into feature vector
output = tokenizer.encode(webpageHtml)
outputDict = collections.Counter(output.ids)
# Apply tfidf weighting
totalFilesUnderConsideration = docDict["totalFilesUnderConsideration"]
array = [0] * tokenizerVocabSize
for item in outputDict:
if len(docDict[item]) > 0:
array[item] = (outputDict[item]) * (math.log10( totalFilesUnderConsideration / len(docDict[item])))
predictionProbability = model.predict_proba([array])[0][1]
print("\n****************************\n--> Probability that the website is phishing: %.2f" % (predictionProbability * 100))
prediction = "NOT PHISHING"
predicted_value = 0
if predictionProbability > threshold:
prediction = "PHISHING"
predicted_value = 1
print("--> Based on your threshold of %.2f, this website is +++'%s'+++" % (threshold, prediction))
print("****************************")
#print(predicted_value)
if predicted_value == 0:
value = "Legitimate"
return render_template("home.html",error=value)
else:
value = "Phishing"
return render_template("home.html",error=value)
class HuggingfaceTokenizerBPE(nn.Module):
def __init__(self, text_files, dataset_info_path='', config_data=None):
super().__init__()
# The default vocab size in the BERT model is 30522. If we want a number larger than that, we will also have to
# change the BERT configuration.
vocab_size = 30000
self.info = f'hug{vocab_size}'
with open(f'config/data/{config_data}.json') as json_file:
tokenizer_from = json.load(json_file)['tokenizer_from']
config_name = config_data if tokenizer_from == "" else tokenizer_from
print(
os.path.join(dataset_info_path,
f'tokenizer_{config_name}_{vocab_size}-vocab.json'))
# The loading is only properly implemented starting from version 0.8. However, it makes the system use a lot of
# CPU for no reason (it is much slower). Maybe it will be fixed in the future.
if not os.path.isfile(
os.path.join(
dataset_info_path,
f'tokenizer_{config_name}_{vocab_size}-vocab.json')):
text_files = text_files()
self.tokenizer = ByteLevelBPETokenizer()
# Join into a single file. This should NOT be necessary but it does not work properly with a lot of files
with open('/tmp/text_files.txt', 'wb') as outfile:
for filename in tqdm(
text_files,
desc='Joining all files into one for tokenization'):
with open(filename, 'rb') as readfile:
shutil.copyfileobj(readfile, outfile)
text_files = '/tmp/text_files.txt'
self.tokenizer.train(text_files,
vocab_size=vocab_size,
special_tokens=special_tokens)
self.tokenizer.save(dataset_info_path,
f'tokenizer_{config_name}_{vocab_size}')
# No "else", always load for consistency
vocab_file = os.path.join(
dataset_info_path,
f'tokenizer_{config_name}_{vocab_size}-vocab.json')
merges_file = os.path.join(
dataset_info_path,
f'tokenizer_{config_name}_{vocab_size}-merges.txt')
self.tokenizer = ByteLevelBPETokenizer(vocab_file=vocab_file,
merges_file=merges_file)
self.tokenizer.add_special_tokens(special_tokens)
self.index_special_tokens = {
tok: self.tokenizer.encode(tok).ids[0]
for tok in special_tokens
}
@property
def device(self):
return self._float_tensor.device
def encode(self, sentence: str):
output = self.tokenizer.encode(sentence)
token_ids = output.ids
tokens = output.tokens
return torch.tensor(token_ids), tokens
def decode(self, tokens: torch.LongTensor):
assert tokens.dim() == 1
tokens = list(tokens.cpu().numpy())
sentences = self.tokenizer.decode(tokens)
return sentences
def id_to_token(self, token_id):
if type(token_id) != torch.Tensor:
token_id = torch.tensor(token_id)
return self.tokenizer.id_to_token(token_id)
def token_to_id(self, token):
assert type(token) == str
return self.tokenizer.token_to_id(token)
def __len__(self):
return self.tokenizer.get_vocab_size()
# This is simply for PyCharm to find the correct reference to the methods of the class
def __call__(self, *input, **kwargs) -> typing.Any:
return super().__call__(*input, **kwargs)
