def addPatron(fn, ln):
conn = sqlite3.connect(Config.getDBFileName())
c = conn.cursor()
fn = fn.upper()
ln = ln.upper()
date = getAmerDate()
record = (fn, ln, date)
c.execute('INSERT INTO vcl VALUES (?,?,?)', record)
conn.commit()
conn.close()
def __init__(self, user, depth=1):
self.user = user
self.depth = depth
self.error_profiles = set()
self.config = Config()
self.profiles = {str(user): {"personaname": "", "friends": []}}
self.db = MySQLdb.connect(host=self.config.db_host,
user=self.config.db_user,
passwd=self.config.db_pw,
db=self.config.db)
def __init__(self, verbose=True):
self.config = Config()
self.verbose = verbose
self.db = MySQLdb.connect(host=self.config.db_host,
user=self.config.db_user,
passwd=self.config.db_pw,
db=self.config.db)
self.q = Queue()
self.db.autocommit(on=True)
self.db.set_character_set('utf8')
with self.db.cursor() as dbc:
dbc.execute('SET NAMES utf8;')
dbc.execute('SET CHARACTER SET utf8;')
dbc.execute('SET character_set_connection=utf8;')
def checkForBan(lnfn):
conn = sqlite3.connect(Config.getDBFileName())
c = conn.cursor()
lnfnarr = nameVAndF(lnfn)
if not lnfnarr:
return ["ERROR: Incorrect formatting."]
ln = lnfnarr[0]
fn = lnfnarr[1]
c.execute('SELECT * FROM banned_patrons WHERE fname=? AND lname=?', (fn, ln))
res = c.fetchall()
conn.close()
if res:
return res[0]
else:
return False
def addBannedPatron(fn, ln, bd, ed, barcode, muni, reason, additional_info):
current_table = getAll('banned_patrons')
if not current_table:
logTC("First record created")
idnum = 1
else:
highest = current_table[0][0]
for record in current_table:
if record[0] > highest:
highest = record[0]
else:
pass
idnum = highest + 1
conn = sqlite3.connect(Config.getDBFileName())
c = conn.cursor()
fn = fn.upper()
ln = ln.upper()
muni = muni.upper()
record = (idnum, fn, ln, bd, ed, barcode, muni, reason, additional_info)
c.execute('INSERT INTO banned_patrons VALUES (?,?,?,?,?,?,?,?,?)', record)
conn.commit()
conn.close()
def __init__(self):
ShowBase.__init__(self)
# Load app configs
self.config = Config(self)
self.debug = False
## EVENT HANDLER ##
self.event = Event(self)
# Event keeper
self.events = {
"player-reset": self.event.doReset,
"change-movement": self.event.doChangeMovement
}
# Accept Events
for eventName in self.events.keys():
self.accept(eventName, self.events[eventName])
# Game.
self.game = Game(self)
self.accept('escape', self.game.stopGame)
def checkForDB():
try:
conn = sqlite3.connect(Config.getDBFileName())
conn.close()
except sqlite3.Error:
return False
mask.append(False)
return mask
def find_wavdir(index):
for (root, subdirs, filenames) in walk(config.train_dir):
for search in subdirs:
path = root + '/' + search
for (_, _, filenames) in walk(path):
for filename in filenames:
if index == filename:
wavdir = path + '/' + index
return wavdir
config = Config('conv')
df = pd.read_csv(config.train_cats + '.csv')
df.set_index('fname', inplace=True)
for f in df.index:
wavdir = find_wavdir(f)
try:
rate, signal, _ = wavfile.read(wavdir)
df.at[f, 'length'] = signal.shape[0] / rate
except:
print(sys.exc_info()[0])
#gives the legnth of the signal in terms of seconds
classes = list(np.unique(df.label))
class_dist = df.groupby(['label'])['length'].mean()
n_samples = 2 * int(df['length'].sum() / 0.1)
prob_dist = class_dist / class_dist.sum()
choices = np.random.choice(class_dist.index, p=prob_dist)
fig, ax = plt.subplots()
ax.set_title('Class Distribution', y=1.08)
ax.pie(class_dist,
labels=class_dist.index,
autopct='%1.1f%%',
shadow=False,
startangle=90)
ax.axis('equal')
plt.show()
config = Config(mode='time')
if config.mode == 'conv':
X, y = build_rand_feat()
y_flat = np.argmax(y, axis=1)
input_shape = (X.shape[1], X.shape[2], 1)
model = get_conv_model()
elif config.mode == 'time':
X, y = build_rand_feat()
y_flat = np.argmax(y, axis=1)
input_shape = (X.shape[1], X.shape[2])
model = get_recurrent_model()
class_weight = compute_class_weight('balanced', np.unique(y_flat), y_flat)
classes = list(np.unique(df.label)) #classes = class names ([burping, hungry,...])
class_dist = df.groupby(['label'])['length'].mean() #average length of each class wav files
#print the class distribution
'''
fig, ax = plt.subplots()
ax.set_title('Class Distribution', y=1.08)
ax.pie(class_dist, labels=class_dist.index, autopct='%1.1f%%',
shadow=False, startangle=90)
ax.axis('equal')
plt.show()
'''
#initialize variables
config = Config(mode='conv') #basic variables
X, y = build_rand_feat() #return random samples from the classes to prevent class inbalance, and return the mfcc version not the time domain
if config.mode == 'conv':
input_shape = (X.shape[1], X.shape[2], 1)
model = get_conv_model()
elif config.mode == 'time':
input_shape = (X.shape[1], X.shape[2])
model = get_recurrent_model()
y_flat = np.argmax(y, axis=1) #return the index corresponding to the 1 value in the y array
classWeight = compute_class_weight('balanced', np.unique(y_flat), y_flat) #A way to modify the hyperparameters a little bit to reduce the bias and improve accuracy a little bit
checkpoint = ModelCheckpoint(config.model_path, monitor='val_acc', verbose=1
, mode='max', save_best_only=True, save_weights_only=False, period=1) #this line makes us update the model we saved before only if there is an acc improvement
Conv2D(16, (3, 3), activation="relu", strides=(1, 1), padding="same"))
model.add(MaxPool2D(2, 2))
model.add(Dropout(0.5))
model.add(Flatten())
model.add(Dense(128, activation="relu"))
model.add(Dense(64, activation="relu"))
model.add(Dense(10, activation="softmax"))
model.summary()
model.compile(loss="categorical_crossentropy",
optimizer="adam",
metrics=["acc"])
return model
#build two directoies one "model" "pickle" (model data)
config = Config(mode="time")
#read training set+label csv
df = pd.read_csv('instruments.csv')
df.set_index('fpath', inplace=True) #set filepath as index
for f in df.index: #
rate, signal = wavfile.read('clean/' + f)
df.at[f, 'length'] = signal.shape[0] / rate
classes = list(np.unique(df.label))
#create a new column of length based on signal length
class_dist = df.groupby(['label'])['length'].mean()
fig, ax = plt.subplots()
ax.set_title('Class Distribution', y=1.08)
import os
import json
from cfg import Config
save_path = './tests/test.json'
cfg = Config()
cfg.save_dir = 'ALEX TESTING'
cfg.batch_size = 144
cfg.save_config(save_path)
with open(save_path, 'r') as f:
saved_dict = json.load(f)
assert saved_dict['save_dir'] == 'ALEX TESTING'
assert saved_dict['batch_size'] == 144
df = pd.read_csv('maestro-v2.0.0.csv')
df['audio_filename'] = df['audio_filename'].str[5:]
df.set_index('audio_filename', inplace=True)
#manually set the classes
classes = ['Baroque', 'Classical', 'Romantic', 'Modern']
classifier = 'period'
class_dist = df.groupby(classifier).period.agg('count').to_frame('countt')
leakyrelu = LeakyReLU(alpha=0.3)
activation_layer = leakyrelu
figsize = 8
#change this for setting machine learning configuration: 'svm', 'time' or 'conv' for mode, and 'chroma' or 'mfcc' for feature
config = Config(mode='svm', feature='chroma')
#for convolutional networks about 6 epochs works best. for LSTMS it is about 9.
epochs = 6
#check if data already exists and open it if yes
def check_data():
if os.path.isfile(config.p_path):
print('Loading existing data for {} model with {} features'.format(
config.mode, config.feature))
with open(config.p_path, 'rb') as handle:
tmp = pickle.load(handle)
return tmp
else:
return None
prob_dist = class_dist / class_dist.sum(
) #this will convert everything between 0 and 1 (probability)
choices = np.random.choice(class_dist.index,
p=prob_dist) #no clue what this does
fig, ax = plt.subplots()
ax.set_title('Class Distribution', y=1.08)
ax.pie(class_dist,
labels=class_dist.index,
autopct='%1.1f%%',
shadow=False,
startangle=90)
ax.axis('equal')
plt.show()
config = Config(
mode='time') #use mode = 'conv' for cnn and mode = 'time' for rnn
#what happens if it runs in convolution
if config.mode == 'conv':
X, y = build_rand_feat(
) #this will form the feature set from the random sampling done above
y_flat = np.argmax(
y, axis=1
) #we are returning the hot encoded index value of labels to the original string labels
input_shape = (X.shape[1], X.shape[2], 1)
model = get_conv_model()
elif config.mode == 'time': #for reccurent neural network
X, y = build_rand_feat()
y_flat = np.argmax(
y, axis=1
import cv2
import numpy as np
from vo import vo
from cfg import Config
from frame import Frame
from matplotlib import pyplot as plt
conf = Config('./cfg/freiburg.json')
vio = vo(conf)
outfile = './out.txt'
x = 1.33
y = 0.64
z = 1.65
# imgs = []
# deps=[]
#
# f=plt.figure()
#
# for i in [1,2]:
# imgs.append(cv2.imread('./data/'+str(i)+'.png'))
# deps.append(cv2.imread('./data/'+str(i)+'d.png', -1))
#
# for i in range(len(imgs)):
# id = int(i)
# fr = Frame(id=id,
# img=imgs[i],
# depth=deps[i],
# camera=vio.camera
# )
# vio.add_frame(fr,visual=True)
from datasets import get_dataloaders from cfg import Config cfg = Config() train_loader, val_loader, test_loader = get_dataloaders(cfg)
return X, y
df = pd.read_csv(
"/content/drive/MyDrive/Audio Signal Processing/Clean Data/instruments_clean.csv"
)
df.set_index('Index', inplace=True)
classes = list(np.unique(df.Label))
class_dist = df.groupby(['Label'])['length'].mean()
n_samples = 2 * int(df['length'].sum() / 0.1)
prob_dist = class_dist / class_dist.sum() #Probability Distribution
choices = np.random.choice(class_dist.index, p=prob_dist)
config = Config(mode='conv')
if config.mode == 'conv':
X, y = build_rand_feat()
y_flat = np.argmax(y, axis=1)
input_shape = (X.shape[1], X.shape[2], 1)
model = get_conv_model()
elif config.mode == 'time':
X, y = build_rand_feat()
y_flat = np.argmax(y, axis=1)
input_shape = (X.shape[1], X.shape[2])
model = get_recurrent_model()
class_weight = compute_class_weight('balanced', np.unique(y_flat), y_flat)
model.fit(X,
y,
n_samples = 2 * int(df['length'].sum() / 0.1)
prob_dist = class_dist / class_dist.sum()
choices = np.random.choice(class_dist.index, p=prob_dist)
fig, ax = plt.subplots()
ax.set_title('Class Distribution', y=1.08)
ax.pie(class_dist,
labels=class_dist.index,
autopct='%1.1f%%',
shadow=False,
startangle=90)
ax.axis('equal')
#plt.show()
config = Config(mode='time') #For selecting between CNN and RNN
if config.mode == 'conv':
X, y = build_rand_feat()
y_flat = np.argmax(y, axis=1)
input_shape = (X.shape[1], X.shape[2], 1)
model = get_conv_model()
elif config.mode == 'time':
X, y = build_rand_feat()
y_flat = np.argmax(y, axis=1)
input_shape = (X.shape[1], X.shape[2])
model = get_recurrent_model()
class_weight = compute_class_weight('balanced', np.unique(y_flat), y_flat)
checkpoint = ModelCheckpoint(config.model_path,