def main():
parser = argparse.ArgumentParser()
parser.add_argument("--input_data_name", type=str, help="Dataset name as registered for use")
parser.add_argument("--dropped_columns", type=str, help="Columns to be dropped. To be inputed one after the other.")
parser.add_argument("--threshold", type=float, help="Percentage of missing values dropped for dropped columns")
parser.add_argument("--output_data", type=str, help="Output cleansed data.")
# KNNImputer args
parser.add_argument("--n_neighbors", type=int, default=5, help="Number of neighbors for KNNImputer")
args = parser.parse_args()
# Get dataset by name
data = run.input_datasets[args.input_data_name]
data_df = data.to_pandas_dataframe()
clean_df = clean_data(data_df, threshold=args.threshold, dropped_columns=literal_eval(args.dropped_columns))
print("Shape of cleaned dataset:\n", clean_df.shape)
# Pop default_status so that it is not included in KNNImputer model
y = clean_df.pop("default_status")
imputer = KNNImputer(n_neighbors=args.n_neighbors, weights="distance", add_indicator=False)
imputed_df = DataFrame(imputer.fit_transform(clean_df), columns=clean_df.columns)
imputed_df["default_status"] = y
print("Fitted KNNImputer. Filled missing values.")
# Dump model artifact
os.makedirs('outputs', exist_ok=True)
joblib.dump(imputer, "outputs/knnimputer.joblib")
print("Saved KNNImputer artifact.")
if not (args.output_data is None):
write_output(imputed_df, path=args.output_data, filename="/cleaned.parquet")
def main():
# Load pre-trained model
model = Classifier(num_chan=num_chan).to(device)
model.load_state_dict(
torch.load(os.path.join(epoch_dir,
str(epoch) + '.model')))
# Create dataset loader over all training set
dataset = testDL(data_dir, 'test')
dataloader = data.DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=True,
num_workers=4)
out_arr = []
fname_arr = []
data_iter = iter(dataloader)
for idx, dat in enumerate(data_iter):
pics = dat[0].to(device).float()
fnames = dat[1]
out = torch.sigmoid(model(pics)).detach().cpu().numpy()
out[out >= threshold] = 1
out[out < threshold] = 0
fname_arr += [*fnames]
out_arr += [*out]
write_output(fname_arr, out_arr, results_fname)
def post(self):
name = self.request.get('alias')
acct = self.request.get('username')
mail = self.request.get('mail')
pw = self.request.get('password')
icon = self.request.get('icon')
status = SUCCESS
json_query_data = ''
if pw == '' or acct == '' or name == '':
status = 502
json_query_data = 'Missing alias, username or password'
else:
if utils.exists(User, [['username', acct]]):
status = 503
json_query_data = 'Already existing account'
else:
# Create the User object and return its key
newuser = User()
newuser.name = name
newuser.username = acct
newuser.mail = mail
newuser.icon = icon
c = Credentials()
c.set_dk(pw)
newuser.password = c
pkey = newuser.put()
json_query_data = str(pkey.id())
utils.write_output(self, json_query_data, status)
def run_inputs(event, input_file, output_file, best_so_far_file):
print("\nBeginning " + input_file)
num_wizards, num_constraints, wizards, constraints = utils.read_input(
input_file)
solution = backtrack_solve(wizards, constraints)
print("\nFound Solution")
print(solution)
utils.write_output(output_file, solution)
def parse_arguments(self, arguments):
input_bytes = read_input(arguments.input, use_bytes=True)
key_bytes = read_input(arguments.key, use_bytes=True)
# process input
xored_bytes = self.xor_bytes(input_bytes, key_bytes)
write_output(xored_bytes, arguments.output, use_bytes=True)
def main():
path_uno = "./asset/bonazzi.it.test.data.txt"
path_due = "./asset/toscano.it.test.data.txt"
reading_uno = utils.read_file(path_uno)
reading_due = utils.read_file(path_due)
eval_uno = utils.extract_word(reading_uno)
eval_due = utils.extract_word(reading_due)
correlation = compute_correlations(init_annotation(eval_uno, eval_due))
utils.write_output(eval_uno, eval_due, correlation)
print(correlation)
def test_anneal():
G = g
result, score = annealing.anneal(G, 120000, 1, 1, 0.0004, print_energy=True)
print(score)
print("C: ", utils.cost_fn(result))
nx.draw(utils.mat_to_nx(G))
plt.savefig("/tmp/g.png")
plt.figure()
nx.draw(utils.mat_to_nx(result))
plt.savefig("/tmp/gres.png")
utils.write_output(result, G, "/tmp/res.txt")
assert utils.verify_in_out(G, "/tmp/res.txt")
def post(self):
json_query_data = ''
status = SUCCESS
mailto = self.request.get('to')
mailsubject = self.request.get('subject')
mailbody = self.request.get('body')
if mailto == '' or mailsubject == '':
json_query_data = 'Missing adress or subject'
status = 510
else:
(status,
json_query_data) = utils.send_email(mailto, mailsubject, mailbody)
utils.write_output(self, json_query_data, status)
def test_initial():
G = utils.read_input("inputs/large-14.in")
print("Running")
plt.figure()
nx.draw(utils.mat_to_nx(G))
plt.savefig("/tmp/G.png")
state = annealing.initial_fn(G)
cost = utils.cost_fn(state)
print("Done with cost ", cost)
plt.figure()
nx.draw(utils.mat_to_nx(state))
plt.savefig("/tmp/tree.png")
utils.write_output(state, G, "/tmp/res.txt")
assert utils.verify_in_out(G, "/tmp/res.txt")
def search(input_path, output_path):
words, grid = parse_input(input_path)
max_row = len(grid)
max_col = len(grid[0])
solution = {}
for word in words:
for r in range(max_row):
for c in range(max_col):
if grid[r][c] == word[0]:
coords = find_word(r, c, max_row, max_col, word, grid)
if coords:
solution[word] = coords
write_output(solution, output_path)
def save_changes():
global routes, geocodes
if routes == None:
setup_input(inputs_plotroutes)
output_filename = datetime.now().strftime(
"output//results_%Y-%m-%d_%H%M%S.csv")
write_output(None, output_filename, routes)
object_output_filename = datetime.now().strftime(
"output//obj_files_%Y-%m-%d_%H%M%S.obj")
saving_objects = open(object_output_filename, "wb")
pickle.dump(routes, saving_objects)
saving_objects.close()
def full_run():
global start_time, start_time_orig, working_on_sped, run_finished
run_finished = False
setup_map_data(constants.FILENAMES[3])
#First, try to find good parameters by doing quick runs that
#don't do improvement procedures or bus assignment.
setup_parameters(constants.FILENAMES[6], True)
working_on_sped = True
start_time = process_time()
start_time_orig = process_time()
print("Searching for good algorithm parameters for special ed.")
vary_params(True, minutes=min(20, constants.MINUTES_PER_SEGMENT / 2))
print("Special ed parameters chosen. Beginning routing")
start_time = process_time()
sped_routes = permutation_approach(True,
minutes=constants.MINUTES_PER_SEGMENT *
3 / 2)
print("Special ed routing finished.")
setup_parameters(constants.FILENAMES[6], False)
working_on_sped = False
start_time = process_time()
start_time_orig = process_time()
print("Searching for good algorithm parameters for magnet routing.")
vary_params(False, minutes=min(20, constants.MINUTES_PER_SEGMENT / 2))
print("Magnet parameters chosen. Beginning routing")
start_time = process_time()
magnet_routes = permutation_approach(
False, minutes=constants.MINUTES_PER_SEGMENT * 3 / 2)
print("Magnet routing finished.")
all_routes = sped_routes + magnet_routes
print("Final number of magnet routes: " + str(len(magnet_routes)))
print("Mean student travel time of magnet routes: " +
str(mstt(magnet_routes)) + " minutes")
print("Final number of special ed routes: " + str(len(sped_routes)))
print("Mean student travel time of special ed routes: " +
str(mstt(sped_routes)) + " minutes")
output_filename = datetime.now().strftime(
"output//results_%Y-%m-%d_%H%M%S.csv")
write_output(constants.FILENAMES[0], output_filename, all_routes)
object_output_filename = datetime.now().strftime(
"output//obj_files_%Y-%m-%d_%H%M%S.obj")
saving_objects = open(object_output_filename, "wb")
pickle.dump(all_routes, saving_objects)
saving_objects.close()
run_finished = True
return all_routes
def main(args) -> None:
"""
This is the main pipe line to analyze barseq counts.
"""
# creating folder to put log file and barcode counts
runner = Run(args) # here we create folder name which is equal to experiment name
make_barseq_directories(runner) # if there is already folder then this will return error massage
# Add file handler
fh = logging.FileHandler(runner.log, mode="w") # creating a log file
fh.setFormatter(logging.Formatter(
"%(asctime)s - %(levelname)s - %(module)s - %(message)s",
datefmt="%Y-%m-%d %H:%M"))
logger.addHandler(fh)
logger.info("***** Starting barseq *****")
# read barcode from fasta files
logger.info(f"Reading in barcodes from {runner.barcodes.name}")
# read barcode
# barcodes = read_barcodes(runner.barcodes) # this is the old script
barcodes = read_barcodes_new(runner.barcodes) # this is the old script
# Process each sequencing file
seq_files_list = sorted(os.listdir(runner.sequences))
for seq_file in seq_files_list:
if not seq_file.endswith(".DS_Store"):
sample = format_filename(seq_file)
logger.info(f"Counting Barcodes in {sample}")
runner.sample_dict[sample+'_F'] = deepcopy(barcodes)
runner.sample_dict[sample+'_R'] = deepcopy(barcodes)
# Change cwd
with Cd(runner.sequences):
count_barcodes(seq_file, runner.sample_dict,[sample+'_F',sample+'_R'])
# Write to output
logger.info(f"Writing results to {runner.path}")
write_output(runner.sample_dict, barcodes, runner)
# Confirm completion of barseq
logger.info("***** barseq is complete! *****")
def main_loop():
model = initialize_ai_model()
while True:
begin = time.time()
try:
input_pdf_files = utils.sensor.scan()
if input_pdf_files:
batch_id = f'{datetime.now().strftime("%Y%m%d%H%M%S")}'
logger.info(
f'Starting batch "{batch_id}" with '
f'{len(input_pdf_files)} files: "{", ".join([file.name for file in input_pdf_files])}"'
)
working_pdf_files = utils.move_to_working_folder(
input_pdf_files)
for file in working_pdf_files:
logger.info(f'Reading "{file.name}"')
try:
result = run_pipeline(model, file)
output_file = utils.move_to_output_folder(
batch_id, file)
utils.write_output(output_file, result)
utils.log_master_record(
file.name, result['SE_output'][-1]['page_number'])
except BaseException as e:
logger.exception(
f'Error while processing file "{file}" due to \n{e}'
)
if not utils.enable_debug:
utils.delete_cache()
os.system(f'chmod -R a+rw {utils.Folder.output}')
logger.info(f'Batch "{batch_id}" finished.')
except BaseException as e:
logger.exception(e)
if utils.enable_debug:
raise
finally:
sleep_time = int(
utils.settings['sensor']['schedule']) - (time.time() - begin)
time.sleep(sleep_time if sleep_time > 0 else 0)
def post(self):
acct = self.request.get('username')
pw = self.request.get('password')
json_query_data = ''
status = SUCCESS
if acct == '' or pw == '':
json_query_data = 'Missing username or password'
status = 506
else:
results = User.query(User.username == acct).fetch(limit=1)
if len(results) == 0:
json_query_data = 'Unexisting account'
status = 505
else:
user = results[0]
c = Credentials()
c.set_dk(pw)
user.password = c
pkey = user.put()
json_query_data = str(pkey.id())
utils.write_output(self, json_query_data, status)
def parse_arguments(self, arguments):
if arguments.lang == 'en':
self.alphabet = Alphabet(en_letters)
elif arguments.lang == 'ru':
self.alphabet = Alphabet(ru_letters)
# read input
input_text = read_input(arguments.input)
# process input
try:
action = getattr(self, arguments.action + '_action')
except AttributeError:
raise NotImplementedError("Action {0} is not implemented. You have"
"to add method {0}_action".format(
arguments.action))
output_text = action(input_text, arguments)
# write to output file
write_output(output_text, arguments.output)
def gen_complete_test_set(tfc_reader):
"""
It generates Complete Test Set for gate having all
positive controls and no constant line, garbage
For NCT and GT library
"""
rev_non_tfc(tfc_reader)
length = get_length()
tp = list(itertools.product([0, 1], repeat=length))
tfc_gates = get_tfc_gates()
output_writter = open("cts.txt", 'w')
output_writter.write(str("||"))
for i in range(0, len(tp)):
pattern = tp[i]
Flag = 0
olist = initalize_output_dict(pattern)
for gate in tfc_gates:
gate = gate.split()
gate_len = len(gate)
# Check if all values are 1 or not
if (all(x == 1 for x in olist.values())):
if gate_len >= 2:
Flag = 1
olist = gate_operation(gate, olist)
if Flag == 1:
write_output(output_writter, olist)
output_writter.seek(0, 0)
output_writter.close()
return
def runfile(infile, outfile, score_to_beat = 1e99):
G = utils.read_input(infile)
# print("Processing", infile)
if "small" in infile:
param = params["small"]
elif "medium" in infile:
param = params["medium"]
elif "large" in infile:
param = params["large"]
else:
print("NOT A RECOGNIZED FILETYPE!!")
param = params["medium"]
iters, ps, pp, scale = param
result, score = annealing.anneal(G, iters, ps, pp, scale, print_energy=False)
# print(f"- {score}")
if score < score_to_beat:
utils.write_output(result, G, outfile)
# assert utils.verify_in_out(G, outfile)
# TODO REMOVE, Just while fixing bugs
# os.remove(infile)
return score
def post(self):
acct = self.request.get('username')
# pw = self.request.get('password')
json_query_data = ''
status = SUCCESS
if acct == '':
json_query_data = 'Missing username'
status = 507
if status == SUCCESS:
results = User.query(User.username == acct).fetch(limit=1)
if len(results) == 0:
json_query_data = 'Unexisting account'
status = 505
if status == SUCCESS:
user = results[0]
mailto = user.mail
if len(mailto) == 0:
json_query_data = 'Account without email. Password cannot be reset'
status = 508
if status == SUCCESS:
pw = utils.id_generator(PASSWORD_DEFAULT_LENGTH)
c = Credentials()
c.set_dk(pw)
name = user.name
user.password = c
pkey = user.put()
json_query_data = str(pkey.id())
with open('html/mail_reset.html', 'r') as myfile:
subject = myfile.read().replace('\n', '')
logging.info(subject)
subject = subject.format(name, acct, pw)
logging.info(subject)
(ret, status) = utils.send_email(mailto, "Password reset", subject,
True)
if status != SUCCESS:
json_query_data = ret
utils.write_output(self, json_query_data, status)
def post(self):
status = SUCCESS
json_query_data = ''
acct = self.request.get('username')
pw = self.request.get('password')
if pw == '' or acct == '':
status = 501
json_query_data = 'Missing username or password'
else:
results = User.query(User.username == acct).fetch(limit=1)
if len(results) > 0:
c = results[0].password
if not c.verify(pw):
status = 502
json_query_data = 'Invalid username or password' # no hints to hackers
else:
json_query_data = utils.json_formatter(results)
else:
status = 502
json_query_data = 'Invalid username or password' # no hints to hackers
time.sleep(3) # hackers can wait...
utils.write_output(self, json_query_data, status)
def post(self):
name = self.request.get('alias')
acct = self.request.get('username')
mail = self.request.get('mail')
icon = self.request.get('icon')
json_query_data = ''
status = SUCCESS
if acct == '' or name == '':
json_query_data = 'Missing alias or username'
status = 504
else:
results = User.query(User.username == acct).fetch(limit=1)
if len(results) == 0:
json_query_data = 'Unexisting account'
status = 505
else:
user = results[0]
user.name = name
user.mail = mail
user.icon = icon
pkey = user.put()
json_query_data = str(pkey.id())
utils.write_output(self, json_query_data, status)
def gen_level_wise_output(tfc_reader):
"""
It generates leve_wise_output for gate having all
positive controls and no constant line, garbage
For NCT and GT library
"""
gen_non_tfc(tfc_reader)
length = get_length()
# tp -- TestPattern
tp = list(itertools.product([0, 1], repeat=length))
tfc_gates = get_tfc_gates()
output_writter = open("lwo.txt", 'w')
for i in range(0, len(tp)):
pattern = tp[i]
output_writter.write(str("||"))
olist = initalize_output_dict(pattern)
# Printing Input Pattern
write_output(output_writter, olist)
for gate in tfc_gates:
gate = gate.split()
olist = gate_operation(gate, olist)
# Printing Level-Wise-Output
write_output(output_writter, olist)
output_writter.write(str("\n"))
output_writter.seek(0, 0)
output_writter.close()
return
#early_stopping = EarlyStopping(monitor='val_rmse', min_delta=0.01, patience=7, verbose=1)
#callbacks_list = [early_stopping]
target = np.array(train.deal_probability)
hist = model.fit(train_cont['cat_d']['cat_data'] + train_cont['cat_d']["parent_data"] +\
train_cont['cat_d']["region_data"] + train_cont['cat_d']["city_data"] + train_cont['cat_d']["image_data"] +\
train_cont['cat_d']["user_data"] + train_cont['cat_d']["day_data"] + train_cont["other_feat"] + \
[padded_words] ,
target,
batch_size=conf.modelling.batch_size,
epochs=3,# conf.modelling.num_epochs
validation_split = 0.1,
shuffle=True,
verbose=2,
# callbacks = callbacks_list
)
preds = model.predict(
test_cont['cat_d']['cat_data'] + test_cont['cat_d']["parent_data"] +\
test_cont['cat_d']["region_data"] + test_cont['cat_d']["city_data"] + test_cont['cat_d']["image_data"] +\
test_cont['cat_d']["user_data"] + test_cont['cat_d']["day_data"] + test_cont["other_feat"] + \
[padded_test] ,
)
from utils import write_output, plot_history
plot_history(hist)
write_output(preds, conf)
def save(self):
self._community['node_pool'] = self._node_pool
self._community['members'] = self._community_members
utils.write_output(self._community, self._write_path)
def main():
# -----------------------------------------------------------------------------------
# Adjustable Parameters
parser = argparse.ArgumentParser()
parser.add_argument(
'--train', action = 'store_true', help = 'training or scoring')
parser.add_argument(
'--inputfile', type = str, help = 'input data file name')
parser.add_argument(
'--outputfile', type = str, help = 'output prediction file name')
args = parser.parse_args()
# directory for the input data and output prediction:
DATA_DIR = 'data'
OUTPUT_DIR = 'output'
# columns used:
CAT_COLS = ['Auction', 'Transmission', 'WheelType', 'Nationality',
'Size', 'TopThreeAmericanName', 'IsOnlineSale']
NUM_COLS = ['VehicleAge', 'VehOdo', 'VehBCost', 'WarrantyCost',
'MMRCurrentAuctionAveragePrice', 'MMRAcquisitionAuctionAveragePrice',
'MMRCurrentAuctionCleanPrice', 'MMRAcquisitionAuctionCleanPrice',
'MMRCurrentRetailAveragePrice', 'MMRAcquisitionRetailAveragePrice',
'MMRCurrentRetailCleanPrice', 'MMRAcquisitonRetailCleanPrice']
DATE_COLS = ['PurchDate']
LABEL_COL = 'IsBadBuy'
IDS_COL = 'RefId'
# current time for computing recency feature
NOW = '2010-12-31'
# modeling step:
# model checkpoint for future scoring
MODEL_DIR = 'model'
CHECKPOINT_XGB = 'xgb.pkl'
CHECKPOINT_PREPROCESS = 'preprocess.pkl'
# parameter that only relevant for training stage and not scoring
if args.train:
# number of cross validation and hyperparameter settings to try
CV = 10
N_ITER = 5
MODEL_RANDOM_STATE = 4321
# train/validation stratified split
VAL_SIZE = 0.1
TEST_SIZE = 0.1
SPLIT_RANDOM_STATE = 1234
# -----------------------------------------------------------------------------------
logger.info('preprocessing')
checkpoint_preprocess = os.path.join(MODEL_DIR, CHECKPOINT_PREPROCESS)
checkpoint_xgb = os.path.join(MODEL_DIR, CHECKPOINT_XGB)
input_path = os.path.join(DATA_DIR, args.inputfile)
if args.train:
data = clean(input_path, NOW, CAT_COLS, NUM_COLS, DATE_COLS, IDS_COL, LABEL_COL)
ids = data[IDS_COL].values
label = data[LABEL_COL].values
data = data.drop([IDS_COL, LABEL_COL], axis = 1)
# train/test split twice to achieve train/validation/test three way split
df_train, df_test, y_train, y_test, ids_train, ids_test = train_test_split(
data, label, ids, test_size = TEST_SIZE,
random_state = SPLIT_RANDOM_STATE, stratify = label)
df_train, df_val, y_train, y_val, ids_train, ids_val = train_test_split(
df_train, y_train, ids_train, test_size = VAL_SIZE,
random_state = SPLIT_RANDOM_STATE, stratify = y_train)
# obtain finalized columns
num_cols_cleaned = list(SortedSet(df_train.columns) - SortedSet(CAT_COLS))
preprocess = Preprocesser(num_cols = num_cols_cleaned, cat_cols = CAT_COLS)
X_train = preprocess.fit_transform(df_train)
X_val = preprocess.transform(df_val)
X_test = preprocess.transform(df_test)
logger.info('modeling')
eval_set = [(X_train, y_train), (X_val, y_val)]
xgb_tuned = build_xgb(N_ITER, CV, MODEL_RANDOM_STATE, eval_set)
xgb_tuned.fit(X_train, y_train)
if not os.path.isdir(MODEL_DIR):
os.mkdir(MODEL_DIR)
dump(preprocess, checkpoint_preprocess)
dump(xgb_tuned, checkpoint_xgb)
# model evaluation metric reporting
y_pred = []
xgb_best = xgb_tuned.best_estimator_
zipped = zip(
('train', 'validation', 'test'),
(X_train, X_val, X_test),
(y_train, y_val, y_test))
for name, X, y in zipped:
xgb_pred = xgb_best.predict_proba(
X, ntree_limit = xgb_best.best_ntree_limit)[:, 1]
score = round(roc_auc_score(y, xgb_pred), 3)
logger.info('{} AUC: {}'.format(name, score))
y_pred.append(xgb_pred)
ids = np.hstack((ids_train, ids_val, ids_test))
y_pred = np.hstack(y_pred)
else:
data = clean(input_path, NOW, CAT_COLS, NUM_COLS, DATE_COLS, IDS_COL)
ids = data[IDS_COL].values
data = data.drop(IDS_COL, axis = 1)
logger.info('scoring')
preprocess = load(checkpoint_preprocess)
xgb_tuned = load(checkpoint_xgb)
X = preprocess.transform(data)
xgb_best = xgb_tuned.best_estimator_
y_pred = xgb_best.predict_proba(
X, ntree_limit = xgb_best.best_ntree_limit)[:, 1]
if not os.path.isdir(OUTPUT_DIR):
os.mkdir(OUTPUT_DIR)
output_path = os.path.join(OUTPUT_DIR, args.outputfile)
write_output(ids, IDS_COL, y_pred, LABEL_COL, output_path)
test_data = random.sample(train_data, int(len(train_data)*.1)) train_data = [post for post in train_data if post not in test_data] train_data = np.array(train_data) # intialize model model = pizza_model.PizzaModel(params) # train #model.train(train_data) # test #predictions = model.test(test_data) # K-fold testing kf = cross_validation.KFold(len(train_data), n_folds=10) predictions = np.zeros(len(train_data)) for train_index, test_index in kf: train_fold, test_fold = train_data[train_index], train_data[test_index] model.train(train_fold) predictions[test_index] = model.test(test_fold) desired = utils.get_labels_from_post_list(train_data) # evaluate if not params.TESTING: #desired = utils.get_labels_from_post_list(test_data) print classification_report(desired, predictions) else: # write output to file utils.write_output(test_data, predictions)
def main():
global cuda
cuda = torch.cuda.is_available()
if cuda:
train_sequence.cuda = cuda
sequence_tagger.cuda = cuda
utils.cuda = cuda
train_sequence_crafted.cuda = cuda
if args.crafted:
this_train_sequence = train_sequence_crafted
else:
this_train_sequence = train_sequence
utils.log('start reading ner file ')
(token_list, tag_list,
raw_token_list) = utils.prepare_data(args.input, True)
vocabs = pickle.load(open(args.vocab_path, 'rb'))
y = list(
map(lambda x: np.array(list(map(lambda y: vocabs['y_dict'][y], x))),
tag_list))
x = tdh.build_input_data(token_list, vocabs['vocabulary'])
#extract crafted features
train_data = utils.get_data_with_pos_tag(raw_token_list, tag_list)
features = utils.extract_features(train_data, vocabs['uptl'],
vocabs['treatment_suffix'],
vocabs['disease_suffix'], vocabs['dis'])
ds_data = {'x': x, 'y': y, 'z': features}
ds = sequence_dataset.sequence_dataset(
'.',
'test',
ds_data,
word_counts=vocabs['word_counts'],
vocabulary_inv=vocabs['vocabulary_inv'],
crafted_features=args.crafted)
val_loader = DataLoader(ds,
batch_sampler=data_samplers.BatchSampler(
list(map(lambda x: min(999999, len(x[0])),
ds)),
256,
shuffle=False),
num_workers=4)
vocab_size = ds.vocab_size
embedding_init = vocabs['embedding_init']
embedding_init = embedding_init[:vocab_size]
if args.model == 'bilstm':
if args.crafted:
model = sequence_tagger.BilstmSequenceTaggerCraftedFeatures(
len(vocabs['y_dict']),
vocab_size,
embedding_size=embedding_init.shape[1],
hidden_size=args.hidden_size,
intermediate_size=args.intermediate_size,
embedding_init=embedding_init,
crafted_features_size=args.num_crafted)
criterion = nn.CrossEntropyLoss()
if cuda:
criterion.cuda()
#
my_loss_fn = lambda x, y, z, m: utils.std_loss_fn_crafted(
x, y, z, m, criterion)
else:
model = sequence_tagger.BilstmSequenceTagger(
len(vocabs['y_dict']),
vocab_size,
embedding_size=embedding_init.shape[1],
hidden_size=args.hidden_size,
intermediate_size=args.intermediate_size,
embedding_init=embedding_init)
criterion = nn.CrossEntropyLoss()
if cuda:
criterion.cuda()
#
my_loss_fn = lambda x, y, m: utils.std_loss_fn(x, y, m, criterion)
else:
model = sequence_tagger.BilstmCRFSequenceTagger(
len(vocabs['y_dict']),
vocab_size,
embedding_size=embedding_init.shape[1],
hidden_size=args.hidden_size,
intermediate_size=args.intermediate_size,
embedding_init=embedding_init)
my_loss_fn = utils.lstm_crf_neg_log_likelihood_loss1
checkpoint = torch.load(args.checkpoint)
model.load_state_dict(checkpoint['model'])
rec, i, all_pred = this_train_sequence.compute_sequence(-1,
model,
my_loss_fn,
val_loader,
None,
'eval',
None,
None, [],
return_preds=True)
utils.write_output(all_pred, raw_token_list, vocabs['y_dict_inv'],
args.output)
# print progress only ten times, expensive because of get_score
if cur_step % (num_steps // 10) == 0:
progress_printer.print(
cur_step, get_score(vehicle_to_rides, bonus, num_steps))
return vehicle_to_rides
def to_string(vehicle_to_rides):
lines = []
for vehicle in vehicle_to_rides:
rides = vehicle_to_rides[vehicle]
lines.extend(
[f"{len(rides)} {' '.join(map(lambda r: str(r.number), rides))}"])
return lines
if __name__ == '__main__':
instances = [
'a_example', 'b_should_be_easy', 'c_no_hurry', 'd_metropolis',
'e_high_bonus'
]
for instance in instances:
print(f'\n\033[95msolving instance {instance}:\033[0m')
rows, columns, num_vehicles, num_rides, bonus, num_steps, rides = utils.read_input(
instance, get_input)
vehicle_to_rides = get_greedy_solution2(num_vehicles, rides, bonus,
num_steps)
utils.write_output(instance, to_string(vehicle_to_rides))
def get_initial_allocation(cache_importance, video_sizes, num_caches, cache_size):
allocation = {c: set() for c in range(num_caches)}
space_left = {c: cache_size for c in range(num_caches)}
already_cached = []
for cache in cache_importance:
for video, _ in cache_importance[cache]:
if video_sizes[video] <= space_left[cache] and video not in already_cached:
already_cached.append(video)
allocation[cache].add(video)
space_left[cache] -= video_sizes[video]
return allocation
def to_string(allocation):
used_caches = {cache for cache in allocation if allocation[cache]}
lines = [f"{len(used_caches)}"]
for cache in allocation:
lines.extend([f"{cache} {' '.join(map(str, allocation[cache]))}"])
return lines
if __name__ == '__main__':
instances = ['example', 'me_at_the_zoo', 'videos_worth_spreading', 'trending_today', 'kittens']
for instance in instances:
print(f'\n\033[95msolving instance {instance}:\033[0m')
latencies, requests, video_sizes, cache_size, num_caches = utils.read_input(instance, get_input)
cache_importance = get_cache_importance(latencies, requests, video_sizes, num_caches)
allocation = solve_by_local_search(cache_importance, latencies, requests, video_sizes, cache_size, num_caches)
utils.write_output(instance, to_string(allocation))
def populate_graph(graph, node1, node2):
if node1 in graph.keys():
graph[node1].append(node2)
else:
graph[node1] = [node2]
return graph
def calc(lines):
parser = re.compile("[A-Z0-9]+")
values = [tuple(parser.findall(line.strip())) for line in lines]
graph = {}
nodes = []
for value in values:
nodes.append(value[1])
graph = populate_graph(graph, value[0], value[1])
you_path = find_path(graph, "YOU", "COM", [])
san_path = find_path(graph, "SAN", "COM", [])
xor = set(you_path) ^ set(san_path)
return len(xor)
if __name__ == '__main__':
lines = read_input()
result = str(calc(lines))
write_output(result)
check_result(result)
def main():
# -----------------------------------------------------------------------------------
# Adjustable Parameters
parser = argparse.ArgumentParser()
parser.add_argument('--train',
action='store_true',
help='training or scoring')
parser.add_argument('--inputfile', type=str, help='input data file name')
parser.add_argument('--outputfile',
type=str,
help='output prediction file name')
args = parser.parse_args()
# directory for the input data and output prediction:
DATA_DIR = 'data'
OUTPUT_DIR = 'output'
# columns used:
CAT_COLS = [
'Auction', 'Transmission', 'WheelType', 'Nationality', 'Size',
'TopThreeAmericanName', 'IsOnlineSale'
]
NUM_COLS = [
'VehicleAge', 'VehOdo', 'VehBCost', 'WarrantyCost',
'MMRCurrentAuctionAveragePrice', 'MMRAcquisitionAuctionAveragePrice',
'MMRCurrentAuctionCleanPrice', 'MMRAcquisitionAuctionCleanPrice',
'MMRCurrentRetailAveragePrice', 'MMRAcquisitionRetailAveragePrice',
'MMRCurrentRetailCleanPrice', 'MMRAcquisitonRetailCleanPrice'
]
DATE_COLS = ['PurchDate']
LABEL_COL = 'IsBadBuy'
IDS_COL = 'RefId'
# current time for computing recency feature
NOW = '2010-12-31'
# modeling step:
# model checkpoint for future scoring
MODEL_DIR = 'model'
CHECKPOINT_PREPROCESS = os.path.join(MODEL_DIR, 'preprocess.pkl')
CHECKPOINT_XGB = os.path.join(MODEL_DIR, 'xgb.pkl')
# parameter that only relevant for training stage and not scoring
if args.train:
# number of cross validation and hyperparameter settings to try
CV = 10
N_ITER = 5
MODEL_RANDOM_STATE = 4321
# train/validation stratified split
VAL_SIZE = 0.1
TEST_SIZE = 0.1
SPLIT_RANDOM_STATE = 1234
# -----------------------------------------------------------------------------------
logger.info('preprocessing')
input_path = os.path.join(DATA_DIR, args.inputfile)
if args.train:
data = clean(input_path, NOW, CAT_COLS, NUM_COLS, DATE_COLS, IDS_COL,
LABEL_COL)
ids = data[IDS_COL].values
label = data[LABEL_COL].values
data = data.drop([IDS_COL, LABEL_COL], axis=1)
# train/test split twice to achieve train/validaion/test three way split
df_train, df_test, y_train, y_test, ids_train, ids_test = train_test_split(
data,
label,
ids,
test_size=TEST_SIZE,
random_state=SPLIT_RANDOM_STATE,
stratify=label)
df_train, df_val, y_train, y_val, ids_train, ids_val = train_test_split(
df_train,
y_train,
ids_train,
test_size=VAL_SIZE,
random_state=SPLIT_RANDOM_STATE,
stratify=y_train)
# obtain finalized columns
num_cols_cleaned = list(
SortedSet(df_train.columns) - SortedSet(CAT_COLS))
preprocess = Preprocesser(num_cols=num_cols_cleaned, cat_cols=CAT_COLS)
X_train = preprocess.fit_transform(df_train)
X_val = preprocess.transform(df_val)
X_test = preprocess.transform(df_test)
logger.info('modeling')
eval_set = [(X_train, y_train), (X_val, y_val)]
xgb_tuned = build_xgb(N_ITER, CV, MODEL_RANDOM_STATE, eval_set)
xgb_tuned.fit(X_train, y_train)
if not os.path.isdir(MODEL_DIR):
os.mkdir(MODEL_DIR)
dump(preprocess, CHECKPOINT_PREPROCESS)
dump(xgb_tuned, CHECKPOINT_XGB)
# model evaluation metric reporting
y_pred = []
xgb_best = xgb_tuned.best_estimator_
zipped = zip(('train', 'validation', 'test'), (X_train, X_val, X_test),
(y_train, y_val, y_test))
for name, X, y in zipped:
xgb_pred = xgb_best.predict_proba(
X, ntree_limit=xgb_best.best_ntree_limit)[:, 1]
score = round(roc_auc_score(y, xgb_pred), 3)
logger.info('{} AUC: {}'.format(name, score))
y_pred.append(xgb_pred)
ids = np.hstack((ids_train, ids_val, ids_test))
y_pred = np.hstack(y_pred)
else:
data = clean(input_path, NOW, CAT_COLS, NUM_COLS, DATE_COLS, IDS_COL)
ids = data[IDS_COL].values
data = data.drop(IDS_COL, axis=1)
logger.info('scoring')
preprocess = load(CHECKPOINT_PREPROCESS)
xgb_tuned = load(CHECKPOINT_XGB)
X = preprocess.transform(data)
xgb_best = xgb_tuned.best_estimator_
y_pred = xgb_best.predict_proba(
X, ntree_limit=xgb_best.best_ntree_limit)[:, 1]
if not os.path.isdir(OUTPUT_DIR):
os.mkdir(OUTPUT_DIR)
output_path = os.path.join(OUTPUT_DIR, args.outputfile)
write_output(ids, IDS_COL, y_pred, LABEL_COL, output_path)
if args.max_level is None:
max_level = class_tree.get_height()
else:
max_level = args.max_level
wstc = WSTC(input_shape=x.shape,
class_tree=class_tree,
max_level=max_level,
sup_source=args.sup_source,
y=y,
vocab_sz=vocab_sz,
word_embedding_dim=word_embedding_dim,
block_thre=args.gamma,
block_level=args.block_level)
total_counts = sum(word_counts[ele] for ele in word_counts)
total_counts -= word_counts[vocabulary_inv_list[0]]
background_array = np.zeros(vocab_sz)
for i in range(1, vocab_sz):
background_array[i] = word_counts[vocabulary_inv[i]] / total_counts
for level in range(max_level):
y_pred = proceed_level(x, sequences, wstc, args, pretrain_epochs,
self_lr, decay, update_interval, delta,
class_tree, level, expand_num, background_array,
max_doc_length, max_sent_length, len_avg,
len_std, beta, alpha, vocabulary_inv,
common_words)
write_output(y_pred, perm, class_tree, './' + args.dataset)
compute_metrics(y_pred, y)
model = CrossEntropyClassifier(nclasses)
history = model.fit(Xtrain, Ytrain, best, lr)
elif classifier == 'svm_ovo':
K = svm_kernel.build_K(Xtrain)
if do_validation:
model = KernelSVMOneVsOneClassifier(nclasses, svm_kernel)
model.fit(Xtrain, Ytrain, C, validation, K=K, check=True)
if do_prediction:
model = KernelSVMOneVsOneClassifier(nclasses, svm_kernel)
model.fit(Xtrain, Ytrain, C, K=K)
elif classifier == 'svm_ova':
K = svm_kernel.build_K(Xtrain)
if do_validation:
model = KernelSVMOneVsAllClassifier(nclasses, svm_kernel)
model.fit(Xtrain, Ytrain, C, validation, K=K, check=True)
if do_prediction:
model = KernelSVMOneVsAllClassifier(nclasses, svm_kernel)
model.fit(Xtrain, Ytrain, C, K=K)
else:
raise Exception("Unknown classifier")
if do_prediction:
print("Predicting on test data")
Ytest = model.predict(Xtest)
write_output(Ytest, 'results/Yte_' + output_suffix + '.csv')
