import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from sklearn.metrics import accuracy_score
from torch.autograd import Variable
from datetime import datetime


# %% Data

skips.append(list(range(70465, 77016)))
data_raw = pd.read_excel('Fruedal.xlsx', skiprows = list(range(1,9096)))
data_raw = data_raw[0:61368]
leak = pd.read_excel('Fruedal.xlsx', sheet_name='Leak')

#%%
M = data_raw.shape[0]
LogTime = list()
LogLeak = list()
for i in range(M):
    list.append(LogTime, datetime.strptime(data_raw['LogDate'][i], "%Y-%m-%d %H:%M:%S.%f"))
    if data_raw['LogDate'][i] in set(leak['LogDate']):
        list.append(LogLeak, 1)
    else:
        list.append(LogLeak, 0)
data_raw['LogTime'] = LogTime
data_raw['LogLeak'] = LogLeak
last_hour = datetime.time(data_raw['LogTime'][5])
data = list()
leaks = list()
log = list()
for i in range(M):
    if datetime.time(data_raw['LogTime'][i]) <= last_hour:
        list.append(data, data_raw[440][i])
        list.append(leaks, data_raw['LogLeak'][i])
        list.append(log, data_raw['LogDate'][i])
data = np.asarray(data)
leaks = np.asarray(leaks)
log = np.asarray(log)
num_days = 7
N = data.shape[0]
num_inputs = int(N/6-(num_days-1)*6)
inputs = np.zeros((num_inputs, num_days*6))
targets = np.zeros(num_inputs)
logDate = list()
for i in range(num_inputs):
    inputs[i] = data[i*6:num_days*6+i*6]
    targets[i] = int(np.max(leaks[num_days*6+i*6-6:num_days*6+i*6]))
    list.append(logDate, log[num_days*6+i*6-1])
logDate = np.asarray(logDate)

t = 0.6
s = (1-t)/2
x_train = inputs[:int(num_inputs*t)]
y_train = targets[:int(num_inputs*t)].reshape(-1,1)
x_val = inputs[int(num_inputs*t):int(num_inputs*(t+s))]
y_val = targets[int(num_inputs*t):int(num_inputs*(t+s))].reshape(-1,1)
x_test = inputs[int(num_inputs*(t+s)):]
y_test = targets[int(num_inputs*(t+s)):].reshape(-1,1)