def __init__(self, state_dim, action_dim, learner_args):
super(PPO, self).__init__()
self.data = []
self.model_path = learner_args['model_path']
self.device = learner_args['device']
hidden_dim=128
self.fc1 = nn.Linear(state_dim,hidden_dim)
# self.fc_pi = nn.Linear(hidden_dim,action_dim)
self.fc_v = nn.Linear(hidden_dim,1)
self.cdt = CDT(learner_args).to(self.device)
self.pi = lambda x: self.cdt.forward(x, LogProb=False)[1]
self.optimizer = optim.Adam(list(self.parameters())+list(self.cdt.parameters()), lr=learning_rate)
def obtenerCDT():
url = 'https://www.davivienda.com/wps/portal/personas/nuevo/personas/aqui_puedo/construir_mi_futuro/inversiones/cdt/a0487af0-343f-4c94-bc79-9943220507a0/!ut/p/z1/pVPJbsIwEP0VeuBo2bEdjI8JiDZUKapSlviCnJBURs1C1tKvr9NDq4JIKnVO9vi9mXkzHijgDopUNupVVipL5Zu--2KyJ9Nn-rS0sTtbYRNZrmvP58w1vAWD298Ab8GJBiBvZvGVgR4JFD3PDhriG__ir-7x3_johlloiL-B_gz6a28THNsBMc5FsMFmXgNEf63LwQT2d63C0OnEkLx-QDfefoDRD_gaUG-ETvSQbF-3jd2IwPELgdtGRS1cp1mR6D_tdRHzUB2gjyU7GChAgFDTBJShKeDSRACzgEWY44BOTfhw1djrPuk9UcfTSVhQhFlaRe8V3LVhslf6XKRRtc-jotQbVY6RSlWosjH68chTrUZ5HR20V7PLqqhVMUrUKK6rusg6SqPBeiOj8lLsxYw3k9tiJaJTJuNOLIkBDTkFQcg44JwSjJGJmEQwT9ba0rMCIvYc5QB_2bRn8mG71t0nvUXbUg!!/dz/d5/L2dBISEvZ0FBIS9nQSEh/'
html = requests.get(url).content
soup = BeautifulSoup(html, 'html.parser')
a = soup.find('a', string = 'Ver tasas vigentes')
linkPDF = a['href']
nombrePDF = 'CDTDavivienda.pdf'
file = Path(nombrePDF)
pdfResponse = requests.get('https://www.davivienda.com' + linkPDF)
file.write_bytes(pdfResponse.content)
#No reconoce el header con los plazos en dias, toca cambiarlo gradualmente
plazos = [30,56,86,116,141,186,356,540,720]
df = tabula.read_pdf(nombrePDF, pages=22, relative_area=True, area=(49.05, 0, 68.5, 100))
info = df.values.tolist()
montoMinimo = 500000
tasas = info[0]
listaCdts =[]
i = 0
for col in tasas:
if i in range(2,11):
tasa = float(col.strip('%'))
cdt = CDT('Davivienda',plazos[i-2],tasa, montoMinimo, None)
listaCdts.append(cdt)
i+=1
os.remove(nombrePDF)
return listaCdts
def obtenerCDT():
#El url del pdf que tiene las tasas
url = 'https://www.bancocajasocial.com/sites/default/files/files/tasascdt.pdf'
#El archivo donde se va a guardar el pdf
nombrePDF = "CDTBancoCajaSocial.pdf"
file = Path(nombrePDF)
#Guardar el pdf
pdfResponse = requests.get(url)
file.write_bytes(pdfResponse.content)
# Lee el area marcada del pdf (top, left, bottom, right) y lo convierte a un dataframe
df = tabula.read_pdf(nombrePDF, relative_area=True, area=(25, 6, 75, 94))
info = df.values.tolist()
montoMin = int(''.join(x for x in info[2][1] if x.isdigit()))
listaCdts = []
i = 0
for fila in info:
if i in range(5, 10):
plazoDias = int(re.search(r'\d+', fila[0]).group())
tasa = float(fila[1].strip('%'))
cdt = CDT('Banco Caja Social', plazoDias, tasa, montoMin, None)
listaCdts.append(cdt)
i += 1
os.remove(nombrePDF)
return listaCdts
def obtenerCDT():
# url del pdf que contiene las tasas
url = 'https://www.bancodeoccidente.com.co/wps/wcm/connect/banco-de-occidente/c24390e8-b984-44de-b427-e59ff152a22c/tasas-personas-2019.pdf?MOD=AJPERES&CVID=mCXPoS8'
# El archivo donde se va a guardar el pdf
nombrePDF = "CDTBancoDeOccidente.pdf"
file = Path(nombrePDF)
# guardar el pdf
pdfResponse = requests.get(url)
file.write_bytes(pdfResponse.content)
# Lee el area marcada del pdf (top, left, bottom, right) y lo convierte a un dataFrame
df = tabula.read_pdf(nombrePDF,
pages=5,
relative_area=True,
area=(19, 72, 34, 94))
info = df.values.tolist()
listaCDTs = []
for i in info:
dias = ""
for j in i:
if str(j) == "nan":
break
else:
if (dias == ""):
jAct = str(j)[:4]
jAct = jAct.strip("D")
jAct = jAct.strip()
dias = jAct
else:
jtasas = j.replace(',', '.')
jtasas = jtasas.split("%")
del jtasas[3]
t = 0
for ts in jtasas:
montoMin = 0
if (t == 0):
montoMin = 1000000
elif (t == 1):
montoMin = 100000000
else:
montoMin = 500000000
cdt = CDT('Banco de Occidente', int(dias), float(ts),
None, int(montoMin))
listaCDTs.append(cdt)
t += 1
return listaCDTs
def obtenerCDT():
url = 'https://www.bbva.com.co/personas/productos/inversion/cdt/tradicional.html#tasas-de-interes-e.a'
html = requests.get(url).content
soup = BeautifulSoup(html, 'html.parser')
tabla = soup.findAll('div', class_='table--value')
filas = []
cdts = []
montoMinimo = 1000000
plazos = [60, 90, 129, 150, 180, 240, 370, 300, 330, 365]
for div in tabla:
a = div.text
b = a.replace('\n', '')
c = b.replace(' ', '')
d = c.replace(' ', '')
if '$' in d:
filas.append(d)
else:
e = d.replace(' ', '')
filas.append(e)
contador = 0
#print(filas)
for i in range(len(filas)):
if filas[i].isdigit():
if 'NOAPLICA' in filas[i + 1]:
rango = 365
tasa = float(filas[i + 2].replace('%', '').replace(',', '.'))
cdt = CDT('BBVA', rango, tasa, None, montoMinimo)
cdts.append(cdt)
else:
rango = plazos[contador]
tasa = float(filas[i + 2].replace('%', '').replace(',', '.'))
cdt = CDT('BBVA', rango, tasa, None, montoMinimo)
cdts.append(cdt)
contador = contador + 1
return cdts
def obtenerCDT():
url = 'https://www.bancopopular.com.co/wps/portal/bancopopular/inicio/informacion-interes/tasas#table-rates-cdt'
html = requests.get(url).content
soup = BeautifulSoup(html, 'html.parser')
tabla = soup.find_all("table", {"class": "simple-table"})[0]
body = tabla.find('tbody')
head = tabla.find('thead')
i = 0
plazos = []
tasas = []
montoMinimo = -1
for child in head.children:
if type(child) is bs4.Tag:
if i == 0:
i += 1
else:
for th in child.children:
if type(th) is bs4.Tag:
plazo = int(re.search(r'\d+', th.string).group())
plazos.append(plazo)
i = 0
for child in body.children:
if type(child) is bs4.Tag:
for td in child.children:
if type(td) is bs4.Tag:
if i == 0:
montoMinimo = int(
td.string.split('a')[0].split('$')[1].strip(
'').replace('.', ''))
i = 1
else:
tasa = float(td.string.strip('%').replace(',', '.'))
tasas.append(tasa)
break
i = 0
listaCdts = []
while i < len(plazos):
cdt = CDT('Banco Popular', plazos[i], tasas[i], montoMinimo, None)
listaCdts.append(cdt)
i += 1
for cdt in listaCdts:
print(
str(cdt.montoInversion) + "_" + str(cdt.plazoMinDias) + "_" +
str(cdt.tasaEA) + "_" + cdt.banco)
return listaCdts
def obtenerCDT():
url = 'https://www.avvillas.com.co/wps/portal/avvillas/banco/banca-personal/productos/ahorro-e-inversion/cdt-av-villas/!ut/p/z1/04_Sj9CPykssy0xPLMnMz0vMAfIjo8zifQIszTwsTQx8LAJ8LAwcQz28PMz8XYwtfYz0w_EpCPYw1Y8iRr8BDuBoQJx-PAqi8Bsfrh-FxwovF29D_AqMHU3wKjAODjEgoAAUSIScWZAbGhphkOmZ6aioCADbOL_N/dz/d5/L2dBISEvZ0FBIS9nQSEh/'
html = requests.get(url).content
soup = BeautifulSoup(html, 'html.parser')
tabla = soup.findAll('table')[1]
filas = []
cdts = []
plazos = [90, 120, 180, 360, 540]
montoMinimo = 10000000
for child in tabla.children:
if type(child) is bs4.Tag:
for tsup in child.children:
if type(tsup) is bs4.Tag:
cols = []
for td in tsup.children:
if type(td) is bs4.Tag:
if td.string != None:
cols.append(td.string)
if len(cols) > 0:
filas.append(cols)
print(filas)
contador = 0
for fila in filas:
if fila == filas[2]:
for col in fila:
if '%' in col:
col1 = col.replace('%', '')
tasa = float(col1)
cdt = CDT('AvVillas', plazos[contador], tasa, None,
montoMinimo)
cdts.append(cdt)
return cdts
def obtenerCDT():
url = 'https://www.itau.co/tasasytarifas'
html = requests.get(url).content
soup = BeautifulSoup(html, 'html.parser')
a = soup.find('a', string='Tasas productos pasivos')
linkPDF = a['href']
nombrePDF = 'CDTItau.pdf'
file = Path(nombrePDF)
pdfResponse = requests.get('https://www.itau.co' + linkPDF)
file.write_bytes(pdfResponse.content)
df = tabula.read_pdf(nombrePDF,
relative_area=True,
area=(76, 0, 96.16, 100))
info = df.values.tolist()
montoMinimo = 1000000
listaCdts = []
i = 0
limite = len(info)
while i < limite:
plazo = int(re.search(r'\d+', info[i + 1][0]).group())
tasa = ''
if isinstance(info[i][2], str):
tasa = float(info[i][2].strip('%'))
else:
tasa = float(info[i + 1][2].strip('%'))
cdt = CDT('Itau', plazo, tasa, montoMinimo, None)
listaCdts.append(cdt)
i += 3
os.remove(nombrePDF)
return listaCdts
def obtenerCDT():
url = 'https://www.bancow.com.co/wp-content/uploads/2019/03/Precios-y-tarifas-vigentes-cdt-marzo-26-2019.pdf'
nombrePDF = "CDTBancoWWB.pdf"
file = Path(nombrePDF)
pdfResponse = requests.get(url)
file.write_bytes(pdfResponse.content)
df = tabula.read_pdf(nombrePDF)
#print(df)
info = df.values.tolist()
print(info)
#rangos = [info[0][1], info[0][2],info[0][3],info[0][4],info[0][5],info[0][6],info[0][7],info[1][8]]
rangos = [60,90,120,180,270,360,540,720]
montoMinimo = 1000000
cdts = []
for fila in info:
filaActual = info.index(fila)
if filaActual == 3:
for col in fila:
colActual = fila.index(col)
monto = info[filaActual][0]
if colActual > 0:
porcentaje = float(info[filaActual][colActual].replace('%', '').replace(',','.'))
rangoo = rangos[colActual-1]
cdt = CDT('WWB', rangoo, porcentaje, None, montoMinimo)
cdts.append(cdt)
return cdts
class PPO(nn.Module):
def __init__(self, state_dim, action_dim, learner_args):
super(PPO, self).__init__()
self.data = []
self.model_path = learner_args['model_path']
self.device = learner_args['device']
hidden_dim=128
self.fc1 = nn.Linear(state_dim,hidden_dim)
# self.fc_pi = nn.Linear(hidden_dim,action_dim)
self.fc_v = nn.Linear(hidden_dim,1)
self.cdt = CDT(learner_args).to(self.device)
self.pi = lambda x: self.cdt.forward(x, LogProb=False)[1]
self.optimizer = optim.Adam(list(self.parameters())+list(self.cdt.parameters()), lr=learning_rate)
def v(self, x):
if isinstance(x, (np.ndarray, np.generic) ):
x = torch.tensor(x)
x = F.relu(self.fc1(x))
v = self.fc_v(x)
return v
def put_data(self, transition):
self.data.append(transition)
def make_batch(self):
s_lst, a_lst, r_lst, s_prime_lst, prob_a_lst, done_lst = [], [], [], [], [], []
for transition in self.data:
s, a, r, s_prime, prob_a, done = transition
s_lst.append(s)
a_lst.append([a])
r_lst.append([r])
s_prime_lst.append(s_prime)
prob_a_lst.append([prob_a])
done_mask = 0 if done else 1
done_lst.append([done_mask])
s,a,r,s_prime,done_mask, prob_a = torch.tensor(s_lst, dtype=torch.float).to(self.device), torch.tensor(a_lst).to(self.device), \
torch.tensor(r_lst).to(self.device), torch.tensor(s_prime_lst, dtype=torch.float).to(self.device), \
torch.tensor(done_lst, dtype=torch.float).to(self.device), torch.tensor(prob_a_lst).to(self.device)
self.data = []
return s, a, r, s_prime, done_mask, prob_a
def train_net(self):
s, a, r, s_prime, done_mask, prob_a = self.make_batch()
for i in range(K_epoch):
td_target = r + gamma * self.v(s_prime) * done_mask
delta = td_target - self.v(s)
delta = delta.detach()
advantage_lst = []
advantage = 0.0
for delta_t in torch.flip(delta, [0]):
advantage = gamma * lmbda * advantage + delta_t[0]
advantage_lst.append([advantage])
advantage_lst.reverse()
advantage = torch.tensor(advantage_lst, dtype=torch.float).to(self.device)
pi = self.pi(s)
pi_a = pi.gather(1,a)
ratio = torch.exp(torch.log(pi_a) - torch.log(prob_a)) # a/b == exp(log(a)-log(b))
surr1 = ratio * advantage
surr2 = torch.clamp(ratio, 1-eps_clip, 1+eps_clip) * advantage
loss = -torch.min(surr1, surr2) + F.smooth_l1_loss(self.v(s) , td_target.detach())
self.optimizer.zero_grad()
loss.mean().backward()
self.optimizer.step()
def choose_action(self, s, Greedy=False):
prob = self.pi(torch.from_numpy(s).unsqueeze(0).float().to(self.device)).squeeze() # make sure input state shape is correct
if Greedy:
a = torch.argmax(prob, dim=-1).item()
return a
else:
m = Categorical(prob)
a = m.sample().item()
return a, prob
def load_model(self, ):
self.load_state_dict(torch.load(self.model_path))
def obtenerCDT():
#url del pdf que contiene las tasas
url = 'https://cdn.agilitycms.com/scotiabank-colombia/Colpatria/pdf/acerca-de/tasas/Tasas-CDT-personas.pdf'
#El archivo donde se va a guardar el pdf
nombrePDF = "CDTColpatria.pdf"
file = Path(nombrePDF)
#guardar el pdf
pdfResponse = requests.get(url)
file.write_bytes(pdfResponse.content)
# Lee el area marcada del pdf (top, left, bottom, right) y lo convierte a un dataFrame
df = tabula.read_pdf(nombrePDF, relative_area=True, area=(25, 6, 75, 94))
#dataFrame convertifo en lista
info = df.values.tolist()
print(info)
listaCDTs = []
plazos = []
for i in info:
entroPlazo = False
montoAct = ""
k = 0
for j in i:
if str(j) == 'MESES' or str(j) == 'RANGOS':
break
elif str(j) == 'PLAZO' or entroPlazo == True:
if str(j) == 'PLAZO':
entroPlazo = True
elif (entroPlazo == True):
jAct = str(j)[0:3]
jAct = jAct.strip('-')
if (str(j).startswith('>')):
jAct = str(j)[-4:]
jAct = jAct.strip()
if jAct != 'nan':
plazos.append(jAct)
else:
if str(j) == '1.000.000 49.999.999':
montoAct = 1000000
continue
elif str(j) == '50.000.000 99.999.999':
montoAct = 50000000
continue
elif str(j) == '100.000.000 499.999.999':
montoAct = 100000000
continue
elif str(j) == '> 500.000.000':
montoAct = 500000000
continue
if str(j) == 'nan':
continue
else:
dias = plazos[k]
jAct = str(j).strip('%')
jAct = jAct.replace(',', '.')
cdt = CDT('Colpatria', int(dias), float(jAct), None,
int(montoAct))
listaCDTs.append(cdt)
k += 1
return listaCDTs
correct = pred.eq(target.view(-1).data).sum()
loss = criterion(output, target.view(-1))
epoch_training_loss_list.append(loss.detach().cpu().data.numpy())
print('Epoch: {:02d} | Batch: {:03d} | CrossEntropy-loss: {:.5f} | Correct: {}/{} | Difference: {}'.format(
epoch, batch_idx, loss.data, correct, output.size()[0], difference))
tree.save_model(model_path = learner_args['model_path'])
writer.add_scalar('Training Loss', np.mean(epoch_training_loss_list), epoch)
writer.add_scalar('Training Feature Difference', np.mean(epoch_feature_difference_list), epoch)
# Testing stage
tree.eval()
correct = 0.
for batch_idx, (data, target) in enumerate(test_loader):
data, target = data.to(device), target.to(device)
batch_size = data.size()[0]
prediction, _, _ = tree.forward(data)
pred = prediction.data.max(1)[1]
correct += pred.eq(target.view(-1).data).sum()
accuracy = 100. * float(correct) / len(test_loader.dataset)
if accuracy > best_testing_acc:
best_testing_acc = accuracy
testing_acc_list.append(accuracy)
writer.add_scalar('Testing Accuracy', accuracy, epoch)
print('\nEpoch: {:02d} | Testing Accuracy: {}/{} ({:.3f}%) | Historical Best: {:.3f}% \n'.format(epoch, correct, len(test_loader.dataset), accuracy, best_testing_acc))
if __name__ == '__main__':
tree = CDT(learner_args).to(device)
train_tree(tree)