import sacred
import torch
import torch.nn as nn
import torch.nn.functional as F
import pandas as pd
import numpy as np
import sys
import os
from sklearn.model_selection import train_test_split
from sklearn import preprocessing
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
from sacred import Experiment
from sacred.observers import FileStorageObserver
from sacred.observers import MongoObserver

from sacred.observers.file_storage import file_storage_option
from sacred.observers.mongo import mongo_db_option



ex = Experiment("s478815",save_git_info=False)
ex.observers.append(FileStorageObserver('experiment'))

#ex.observers.append(MongoObserver(url='mongodb://mongo_user:mongo_password@localhost:27017', db_name='sacred'))  

@ex.config
def my_config():
    epochs = 1000

# Model
class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(1,1) 

    def forward(self, x):
        y_predicted = torch.sigmoid(self.linear(x))
        return y_predicted

data = pd.read_csv('data.csv')
data.dropna()

training_data = data.sample(frac=0.9, random_state=25)
testing_data = data.drop(training_data.index)

print(f"No. of training examples: {training_data.shape[0]}")
print(f"No. of testing examples: {testing_data.shape[0]}")

training_data = training_data[['sqft_living', 'price']]

testing_data = testing_data[['sqft_living', 'price']]

training_data[['price']] = training_data[['price']] / 10000000
training_data[['sqft_living']] = training_data[['sqft_living']] / 10000

testing_data[['price']] = testing_data[['price']] / 10000000
testing_data[['sqft_living']] = testing_data[['sqft_living']] / 10000

# Tensory
X_training = training_data[['sqft_living']].to_numpy()
X_testing = testing_data[['sqft_living']].to_numpy()
y_training = training_data[['price']].to_numpy()
y_testing = testing_data[['price']].to_numpy()

torch.from_file
X_training = torch.from_numpy(X_training.astype(np.float32))
X_testing = torch.from_numpy(X_testing.astype(np.float32))
y_training = torch.from_numpy(y_training.astype(np.float32))
y_testing = torch.from_numpy(y_testing.astype(np.float32))


model = Model()
criterion = nn.BCELoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

@ex.automain
def my_main(epochs, _run):
# Trening
#num_epochs = EPOCHS
    for epochs in range(epochs):
        y_predicted = model(X_training)
        loss = criterion(y_predicted,y_training)
        loss.backward()
        optimizer.step()
        optimizer.zero_grad()

        if (epochs%100==0):
            print(f'epoch:{epochs+1},loss = {loss.item():.4f}')
            
        with torch.no_grad():
            y_predicted = model(X_testing)
            y_predicted_cls = y_predicted.round()
            acc = y_predicted_cls.eq(y_testing).sum()/float(y_testing.shape[0])
            print(f'{acc:.4f}')
            result = open("output",'w+')
            result.write(f'{y_predicted}')

        torch.save(model, "modelP.pkl")
        
        _run.info['accuracy'] = accuracy_score(y_testing, np.argmax(y_predicted, axis=1))
        
        
ex.run()