def __init__(self):
self.test = utils.fix_path(FLAGS.test_dir)
self.train = utils.fix_path(FLAGS.train_dir)
self.val = utils.fix_path(FLAGS.val_dir)
self.shuffle_buffer = 1000
self.to_yuv = FLAGS.to_yuv
self.input_shape = tuple(FLAGS.train_crop)
assert os.path.exists(self.train), \
"Path train_dir: {} does not exist".format(self.train)
assert os.path.exists(self.val), \
"Path val_dir: {} does not exist".format(self.val)
def parse_config(FILENAME):
scpr = SafeConfigParser()
scpr.read(utils.fix_path(FILENAME))
SMTP_SERVER = scpr.get("smtp", "server")
SMTP_PORT = scpr.get("smtp", "port")
USERNAME = scpr.get('account', 'username')
PASSWORD = scpr.get('account', 'password')
CSVPATH = utils.fix_path(scpr.get("file", "mails_data"))
return {'user': USERNAME,
'passwd': PASSWORD,
'server': SMTP_SERVER,
'port': SMTP_PORT,
'csv_path': CSVPATH}
def train_model(request_dict: dict = None):
"""
train model among options specified in project_conf.json.
:param request_dict: request posted via API
:return: mae, after saving updated model
"""
model = None
if request_dict:
data = pd.DataFrame(request_dict["bitcoin_last_minute"], index=[0])
else:
logging.info("Train mode.")
model_name = conf_object.project_conf["model"]
if model_name == 'rfregressor':
from models.rfregressor import RFregressor
model = RFregressor()
if model_name == 'neuralnet':
from models.neural_net import NeuralNet
model = NeuralNet(data=data)
if model_name == 'lstm':
from models.lstm import LSTM
model = LSTM(data=data)
mae = model.eval()
# save model
with open(os.path.join(fix_path(), 'models/model.pkl'), 'wb') as f:
pickle.dump(model, f)
return mae
def main():
if fix_path(FLAGS.data_set_path, is_folder=False):
data_set = get_data_set()
data_set.save(FLAGS.data_set_path)
else:
logging.debug('didn\'t create dataset')
def create_table(csvfile, db_uri):
engine = create_engine(db_uri, echo=True)
conn = engine.connect()
trans = conn.begin()
db.metadata.create_all(engine)
with open(fix_path(base_path, csvfile), mode="r") as p:
csv_data = csv.DictReader(p)
if csvfile.endswith("yearlytempavg1800.csv"):
try:
for row in csv_data:
conn.execute(Temperature.__table__.insert(), row)
trans.commit()
except:
trans.rollback()
raise
else:
try:
for row in csv_data:
row["Date"] = datetime.strptime(row["Date"], "%Y-%m-%d")
conn.execute(Location.__table__.insert(), row)
trans.commit()
except Exception as e:
print(e)
trans.rollback()
raise
class DevelopmentConfig(Config):
DEBUG = 1
FLASK_DEBUG = 1
SQLALCHEMY_ECHO = 1
TEMPLATES_AUTO_RELOAD = 1
SQLALCHEMY_DATABASE_URI = os.getenv(
"DEV_DATABASE_URI", "sqlite:///" +
fix_path(basedir, "db/belly_button_biodiversity.sqlite"))
def main():
LOG_PATH = "birthreminder.log"
logging.basicConfig(
level=logging.DEBUG,
filename=(utils.fix_path(LOG_PATH)),
format="%(asctime)s %(name)s %(levelname)s %(message)s")
today = datetime.today()
stoday = ddmm_from(today)
log.info("Checking %s ..." % stoday + "...")
ctd, ctmr = check_birthday(today)
log.info("Checked %s , %d today - %d tomorrow" % (stoday, ctd, ctmr))
send_mail(["*****@*****.**"], "Checked", ddmm_from(datetime.now()))
def main():
LOG_PATH = "birthreminder.log"
logging.basicConfig(
level=logging.DEBUG,
filename=(utils.fix_path(LOG_PATH)),
format="%(asctime)s %(name)s %(levelname)s %(message)s"
)
today = datetime.today()
stoday = ddmm_from(today)
log.info("Checking %s ..." % stoday + "...")
ctd, ctmr = check_birthday(today)
log.info("Checked %s , %d today - %d tomorrow" % (stoday, ctd, ctmr))
send_mail(["*****@*****.**"],
"Checked", ddmm_from(datetime.now()))
def predict_api():
try:
model = pd.read_pickle(os.path.join(fix_path(), "models/model.pkl"))
logging.info("RFregressor version: ",
pkg_resources.get_distribution("scikit-learn"))
# observation = observation.encode() # this code is for scenario where data is encoded as str in POST
# observation = pickle.loads(base64.b64decode(observation))
# request = open('request.json', 'rb') # todo - comment out if not testing locally
observation = request.json
observation = process_request(observation=observation)
pred = model.get_prediction(observation)
return jsonify({"bitcoin prediction": str(pred)})
except Exception as ex:
logging.error("No model was found, so run /train")
from tkinter import *
import time
import gaugelib
import utils
# region static for recurrent function
g_value = 0
x = 0
# z = 0
store_data = []
# endregion
# region Excel path fixing
file_path = utils.fix_path()
# endregion
def read_serial_and_save():
global x
g_value = utils.read_arduino_serial()
global store_data
store_data += [g_value]
p1.set_value(int(g_value))
x += 1
if x > 1000:
x = 0
win.after(50, read_serial_and_save)
# utils.save_to_excel(store_data, str(file_path)) # for automatic save
class DevelopmentConfig(Config):
DEBUG = 1
FLASK_DEBUG = 1
SQLALCHEMY_ECHO = 1
TEMPLATES_AUTO_RELOAD = 1
SQLALCHEMY_DATABASE_URI = os.getenv("DEV_DATABASE_URI", "sqlite:///" + fix_path(basedir, "db/final-dev.sqlite"))
class ProductionConfig(Config):
LOG_TO_STDOUT = 1
SQLALCHEMY_DATABASE_URI = os.getenv("DATABASE_URI", "sqlite:///" + fix_path(basedir, "db/yearlyTempAvg1800.sqlite"))
SQLALCHEMY_BINDS = {
"finalDB": "sqlite:///" + fix_path(basedir, "db/final.sqlite"),
}
class ProductionConfig(Config):
LOG_TO_STDOUT = 1
SQLALCHEMY_DATABASE_URI = os.getenv(
"DATABASE_URI", "sqlite:///" +
fix_path(basedir, "db/belly_button_biodiversity.sqlite"))
def main(argv):
tf.random.set_seed(FLAGS.seed)
if FLAGS.tbdir is not None:
summary_writers = utils.create_summary_writers(
utils.fix_path(FLAGS.tbdir))
# prepare dataset
dataset = datasets.get_dataset()()
input_shape = dataset.get_input_shape()
# Create Nets and Optimizers
encoder_decoder = nets.encoder_decoder(
input_shape=input_shape,
msg_length=FLAGS.msg_length,
noise_layers=FLAGS.noise_layers,
n_convbnrelu_encoder=FLAGS.n_convbnrelu_encoder,
n_convbnrelu_decoder=FLAGS.n_convbnrelu_decoder)
discriminator = nets.discriminator(
input_shape=input_shape, n_convbnrelu=FLAGS.n_convbnrelu_discriminator)
optimizer_encoder_decoder = tf.keras.optimizers.Adam(1e-3)
optimizer_discriminator = tf.keras.optimizers.Adam(1e-3)
# global step / epoch variables
step = tf.Variable(0, dtype=tf.int64)
epoch = tf.Variable(0, dtype=tf.int64)
# prepare checkpointer
ckpt = tf.train.Checkpoint(
step=step,
epoch=epoch,
optimizer_encoder_decoder=optimizer_encoder_decoder,
optimizer_discriminator=optimizer_discriminator,
encoder_decoder=encoder_decoder,
discriminator=discriminator)
ckpt_manager = tf.train.CheckpointManager(ckpt,
utils.fix_path(FLAGS.ckptdir),
max_to_keep=FLAGS.keep_ckpts)
if ckpt_manager.latest_checkpoint is not None:
if FLAGS.load_from_ckpt:
ckpt.restore(ckpt_manager.latest_checkpoint)
logging.info("Loading model from checkpoint: {}".format(
ckpt_manager.latest_checkpoint))
# Metrics Tracker
metrics_train = metrics.MetricsTracker()
metrics_val = metrics.MetricsTracker()
while epoch < FLAGS.epochs:
dataset_train = dataset.create_train_dataset()
for epoch_step, cover_images in enumerate(dataset_train):
messages = tf.random.uniform([FLAGS.batch_size, FLAGS.msg_length],
minval=0,
maxval=2,
dtype=tf.int32)
messages = tf.cast(messages, dtype=tf.float32)
time_start = time.time()
outputs = steps.train(
cover_images=cover_images,
messages=messages,
encoder_decoder=encoder_decoder,
discriminator=discriminator,
training=True,
optimizer_encoder_decoder=optimizer_encoder_decoder,
optimizer_discriminator=optimizer_discriminator)
ms_per_step = (time.time() - time_start) * 1000.0
ms_per_sample = ms_per_step / FLAGS.batch_size
# Write step summaries
is_summary_step = (step.numpy() % FLAGS.summary_freq) == 0
if is_summary_step:
step_losses = losses.step_loss(
cover_images,
messages,
encoder_decoder_output=outputs['encoder_decoder'],
discriminator_on_cover=outputs['discriminator_on_cover'],
discriminator_on_encoded=outputs[
'discriminator_on_encoded'])
metrics_train.update(
step_losses,
messages,
encoder_decoder_output=outputs['encoder_decoder'],
discriminator_on_cover=outputs['discriminator_on_cover'],
discriminator_on_encoded=outputs[
'discriminator_on_encoded'])
metrics_train_results = metrics_train.results()
metrics_train.reset()
with summary_writers['train'].as_default():
for _name, _value in metrics_train_results.items():
tf.summary.scalar(_name, _value, step=step)
tf.summary.scalar('ms_per_step', ms_per_step, step=step)
tf.summary.scalar('ms_per_sample',
ms_per_sample,
step=step)
step.assign_add(1)
ckpt_save_path = ckpt_manager.save()
logging.info("Saved model after epoch {} to {}".format(
epoch.numpy(), ckpt_save_path))
# Training Loss
logging.info("Epoch {} Stats".format(epoch.numpy()))
logging.info("Training Stats ===========================")
for _name, _value in metrics_train_results.items():
logging.info("{}: {:.4f}".format(_name, _value))
# Evaluate
dataset_val = dataset.create_val_dataset()
for cover_images in dataset_val:
messages = utils.create_messages(batch_size=cover_images.shape[0],
msg_length=FLAGS.msg_length)
# messages = tf.random.uniform(
# [FLAGS.batch_size, FLAGS.msg_length],
# minval=0, maxval=2, dtype=tf.int32)
# messages = tf.cast(messages, dtype=tf.float32)
outputs = steps.train(cover_images=cover_images,
messages=messages,
encoder_decoder=encoder_decoder,
discriminator=discriminator,
training=False)
losses_val_step = losses.step_loss(
cover_images,
messages,
encoder_decoder_output=outputs['encoder_decoder'],
discriminator_on_cover=outputs['discriminator_on_cover'],
discriminator_on_encoded=outputs['discriminator_on_encoded'])
metrics_val.update(
losses_val_step,
messages,
encoder_decoder_output=outputs['encoder_decoder'],
discriminator_on_cover=outputs['discriminator_on_cover'],
discriminator_on_encoded=outputs['discriminator_on_encoded'])
metrics_val_results = metrics_val.results()
metrics_val.reset()
logging.info("Validation Stats ===========================")
with summary_writers['val'].as_default():
for _name, _value in metrics_val_results.items():
tf.summary.scalar(_name, _value, step=step)
logging.info("{}: {:.4f}".format(_name, _value))
messages = utils.create_messages(batch_size=cover_images.shape[0],
msg_length=FLAGS.msg_length)
encoder_decoder_output = encoder_decoder(inputs={
'cover_image': cover_images,
'message': messages
},
training=False)
# write example images to Summaries
with summary_writers['val'].as_default():
transform_fn = None
if FLAGS.to_yuv:
transform_fn = tf.image.yuv_to_rgb
utils.summary_images(
cover=cover_images,
encoded=encoder_decoder_output['encoded_image'],
transmitted_encoded=encoder_decoder_output[
'transmitted_encoded_image'],
transmitted_cover=encoder_decoder_output[
'transmitted_cover_image'],
step=step,
transform_fn=transform_fn)
epoch.assign_add(1)
def fix_paths():
fix_path(FLAGS.checkpoint_path)