def handler(event, context):
return console_logger('handler 2!')
import asyncio
import time
from flags import FLAGS
from pywizlight.bulb import PilotBuilder, wizlight
import re
from helpers import console_logger
LOGGER = console_logger(__name__, "DEBUG")
_SEQUENCE_NOT_FOUND = "Sekvence hodnot nenalezena. Ignoruji příkaz."
class WizLight:
def __init__(self, ip: str, loop):
self.light = wizlight(ip)
self.loop = loop
self.initial_state = self.get_state()
def set_brightness(self, brightness=None):
if brightness is None:
state = self.get_state()
brightness = state.get_brightness()
brightness = int(brightness)
brightness = max(brightness, 0) # lower bound
brightness = min(brightness, 255) # upper bound
self.loop.run_until_complete(
self.light.turn_on(PilotBuilder(brightness=brightness)))
def set_color(self, rgb=(0, 0, 0)):
import time
import numpy as np
import tensorflow as tf
from helpers import console_logger
_NUM_SAMPLES = 2
_DATA_SHAPE = (10, 5) # time, num_features
_BATCH_SIZE = 1
_NUM_INSTANCES = 1 # > 1 not supported
_BANDWIDTH_RANGE = (2, 5)
_AXIS = 1
LOGGER = console_logger('tensorflow', "INFO")
# TODO:
#
# DONE:
# - TimeMasking
# - FrequencyMasking
class ArtificialDataset(tf.data.Dataset):
def _generator(num_samples):
# Opening the file
time.sleep(0.03)
for sample_idx in range(num_samples):
# Reading data (line, record) from the file
time.sleep(0.015)
def train_model(run_number):
logger = console_logger(__name__, FLAGS.logger_level)
logger.info("Clearning Keras session.")
K.clear_session()
save_path = create_save_path()
logger.debug(f"New save path: {save_path}")
# __INPUT PIPELINE__ #
logger.info("Loading datasets.")
ds_train, ds_test, num_train_batches, num_test_batches = load_datasets(
FLAGS.load_dir)
logger.debug(
f"num_train_batches: {num_train_batches}, num_test_batches: {num_test_batches}"
)
# __MODEL__ #
logger.info("Initializing kernel.")
kernel_initializer = tf.initializers.TruncatedNormal(
mean=FLAGS.weight_init_mean, stddev=FLAGS.weight_init_stddev)
logger.info("Building model")
model = build_model(kernel_initializer)
# print('Trainable params: {}'.format(model.count_params()))
logger.info(model.summary())
# Load model weights from checkpoint if checkpoint_path is provided
if FLAGS.checkpoint_path:
logger.info("Loading model weights from checkpoint.")
model.load_weights(FLAGS.checkpoint_path)
# save model FLAGS to save_path
logger.info(f"Saving flags (settings) to {save_path}")
save_config(save_path)
# save model architecture image to save_path directory
if FLAGS.save_architecture_image:
logger.info(f"Saving model architecture image to {save_path}")
tf.keras.utils.plot_model(model,
os.path.join(save_path, 'architecture.png'),
show_shapes=FLAGS.show_shapes)
# __LOGGING__ #
logger.info(f"Initializing summyary writers.")
summary_writer_train = tf.summary.create_file_writer(save_path + '/train',
name='sw-train')
summary_writer_test = tf.summary.create_file_writer(save_path + '/test',
name='sw-test')
# __TRAINING__ #
logger.info(f"Initializing variables, optimizer and metrics.")
best_cer = 100.0
best_epoch = -1
optimizer = tf.keras.optimizers.Adam(lr=FLAGS.lr,
epsilon=FLAGS.epsilon,
amsgrad=FLAGS.amsgrad)
train_loss = tf.keras.metrics.Mean(name='train_loss', dtype=tf.float32)
test_loss = tf.keras.metrics.Mean(name='test_loss', dtype=tf.float32)
test_cer = tf.keras.metrics.Mean(name='test_cer', dtype=tf.float32)
test_ccer = tf.keras.metrics.Mean(name="test_ccer", dtype=tf.float32)
for epoch in range(FLAGS.max_epochs):
logger.log(35, f'_______| Run {run_number} | Epoch {epoch} |_______')
# TRAINING DATA
with summary_writer_train.as_default():
logger.info(f"Training model.")
train_fn(model, ds_train, optimizer, train_loss, num_train_batches,
epoch)
# TESTING DATA
with summary_writer_test.as_default():
logger.info(f"Testing model.")
_, test_mean_cer = test_fn(model, ds_test, test_loss, test_cer,
test_ccer, num_test_batches, epoch)
# EARLY STOPPING AND KEEPING THE BEST MODEL
stop_training, best_cer, best_epoch = early_stopping(
model, test_mean_cer, best_cer, epoch, best_epoch, save_path)
logger.log(35, f'| Best CER {best_cer} | Best epoch {best_epoch} |')
if stop_training:
logger.log(35, f'Model stopped early at epoch {epoch}')
break
class FLAGS:
# noinspection DuplicatedCode
use_digits = False
if use_digits:
c2n_map = {
'0': 0,
'1': 1,
'2': 2,
'3': 3,
'4': 4,
'5': 5,
'6': 6,
'7': 7,
'8': 8,
'9': 9,
'a': 10,
'á': 11,
'b': 12,
'c': 13,
'č': 14,
'd': 15,
'ď': 16,
'e': 17,
'é': 18,
'ě': 19,
'f': 20,
'g': 21,
'h': 22,
'ch': 23,
'i': 24,
'í': 25,
'j': 26,
'k': 27,
'l': 28,
'm': 29,
'n': 30,
'ň': 31,
'o': 32,
'ó': 33,
'p': 34,
'q': 35,
'r': 36,
'ř': 37,
's': 38,
'š': 39,
't': 40,
'ť': 41,
'u': 42,
'ú': 43,
'ů': 44,
'v': 45,
'w': 46,
'x': 47,
'y': 48,
'ý': 49,
'z': 50,
'ž': 51,
' ': 52
}
else:
c2n_map = {
'a': 0,
'á': 1,
'b': 2,
'c': 3,
'č': 4,
'd': 5,
'ď': 6,
'e': 7,
'é': 8,
'ě': 9,
'f': 10,
'g': 11,
'h': 12,
'ch': 13,
'i': 14,
'í': 15,
'j': 16,
'k': 17,
'l': 18,
'm': 19,
'n': 20,
'ň': 21,
'o': 22,
'ó': 23,
'p': 24,
'q': 25,
'r': 26,
'ř': 27,
's': 28,
'š': 29,
't': 30,
'ť': 31,
'u': 32,
'ú': 33,
'ů': 34,
'v': 35,
'w': 36,
'x': 37,
'y': 38,
'ý': 39,
'z': 40,
'ž': 41,
' ': 42
}
n2c_map = {val: idx for idx, val in c2n_map.items()}
alphabet_size = len(c2n_map)
logger_level = "INFO"
logger = console_logger(__name__, logger_level)
load_dir = "b:/!temp/PDTSC_MFSC_unigram_40_banks_DEBUG_min_100_max_3000_tfrecord/1.0/"
# load_dir = "g:/datasets/PDTSC_Debug/"
# load_dir = "g:/datasets/PDTSC_MFSC_unigram_40_banks_min_100_max_3000_tfrecord/"
# load_dir = "g:/datasets/PDTSC_MFSC_unigram_40_banks_min_100_max_3000_tfrecord_DAspeed/"
# load_dir = "g:/datasets/ORAL_MFSC_unigram_40_banks_min_100_max_3000_tfrecord/"
# load_dir = "g:/datasets/COMBINED_MFSC_unigram_40_banks_min_100_max_3000_tfrecord/1.0/"
save_dir = "./results/"
save_config_as = "FLAGS.py"
checkpoint_path = None
# Tried to remove LM, see what it does now
# Removing LM didn't help so its somewhere else. Removed softmax from final activations.
# Next - Trying to add LM without BN and DP
# Changed lr from 0.01 to 0.1
# TODO:
# Add BN and DP to LM
num_runs = 2
max_epochs = 60
batch_size_per_GPU = 8
fs = 16000 # sampling rate of the loaded audiofiles
# noinspection DuplicatedCode
try:
with open(load_dir + "data_config.json", "r") as f:
dc = json.load(f)
num_train_data = dc[
"num_train_data"] # int(48812/2) # int(11308/2) # full ORAL == 374714/2
num_test_data = dc[
"num_test_data"] # int(4796/2) # int(1304/2) # full ORAL == 16502/2
num_features = dc["num_features"] # 123
min_time = dc["min_time"] # 100
max_time = dc["max_time"] # 3000
except FileNotFoundError:
logger.warning(
f"data_config.json file not found at {load_dir}. Loading default values."
)
num_train_data = 24406
num_test_data = 2398
num_features = 123
min_time = 100
max_time = 3000
buffer_size = int(0.1 * num_train_data / batch_size_per_GPU)
shuffle_seed = 42
bucket_width = 100
feature_pad_val = 0.0
label_pad_val = -1
# MODEL
save_architecture_image = False
show_shapes = True
weight_init_mean = 0.0
weight_init_stddev = 0.0001
ff_first_params = {
'use': False,
'num_units': [128, 64],
'batch_norm': False,
'drop_rates': [0.],
}
conv_params = {
'use': True,
'channels': [32, 64, 128, 256],
'kernels': [(16, 32), (8, 16), (4, 8), (4, 4)],
'strides': [(2, 4), (2, 4), (1, 2), (1, 2)],
'dilation_rates': [(1, 1), (1, 1), (1, 1), (1, 1)],
'padding': 'same',
'data_format': 'channels_last',
'batch_norm': True,
'drop_rates': [0., 0., 0., 0.],
}
bn_momentum = 0.9
relu_clip_val = 20
relu_alpha = 0.2
rnn_params = {
'use': True,
'num_units': [512, 512],
'batch_norm': True,
'drop_rates': [0., 0.],
}
ff_params = {
'use': True,
'num_units': [256, 128],
'batch_norm': True,
'drop_rates': [0., 0.],
}
# LANGUAGE MODEL params
lm_gru_params = {
'use': True,
'num_units': [128, 64],
'batch_norm': False,
'drop_rates': [0.0, 0.0]
}
# Optimizer
lr = 0.01
lr_decay = True
lr_decay_rate = 0.9
lr_decay_epochs = 2
epsilon = 0.1
amsgrad = True
# Data Augmentation (in pipeline)
data_aug = {
'mode':
"0x", # mode of many times to apply data aug (allowed: 0x, 1x or 2x)
'bandwidth_time': (10, 100),
'bandwidth_freq': (10, 30),
'max_percent_time': 0.2,
'max_percent_freq': 1.,
}
# Test Pipeline:
spell_check = True
transcribe_digits = True
# Decoder
beam_width = 32
top_paths = 1 # > 1 not implemented
# Early stopping
patience_epochs = 3
