Desert_Rock_3hour.py

# coding: utf-8

# ### NOTE
# ** the test['Kt'].max() values after grouping into hourly data is 7279.907547 = this is a outlier. The reason might be having a dw_solar reading with QC=2 value. Still haven't checked that. But the second highest value of 'Kt' parameter in the testset is 672 (which is way lower) **

# In[1]:

import numpy as np
import pandas as pd
import datetime
import glob
import os.path
from pandas.compat import StringIO


# ### NREL Bird Model implementation: for obtaining clear sky GHI

# In[2]:

import itertools
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns


# In[3]:

#get_ipython().magic('matplotlib inline')
sns.set_color_codes()


# In[4]:

import pvlib
from pvlib import clearsky, atmosphere
from pvlib.location import Location


# In[5]:

drk = Location(36.621,-116.043, 'US/Pacific', 1010.1072, 'Desert Rock')


# In[6]:

times2009 = pd.DatetimeIndex(start='2009-01-01', end='2010-01-01', freq='1min',
                        tz=drk.tz)   # 12 months of 2009 - For testing
times2010and2011 = pd.DatetimeIndex(start='2010-01-01', end='2012-01-01', freq='1min',
                        tz=drk.tz)   # 24 months of 2010 and 2011 - For training


# In[7]:

cs_2009 = drk.get_clearsky(times2009) 
cs_2010and2011 = drk.get_clearsky(times2010and2011) # ineichen with climatology table by default
#cs_2011 = bvl.get_clearsky(times2011) 


# In[8]:

cs_2009.drop(['dni','dhi'],axis=1, inplace=True) #updating the same dataframe by dropping two columns
cs_2010and2011.drop(['dni','dhi'],axis=1, inplace=True) #updating the same dataframe by dropping two columns
#cs_2011.drop(['dni','dhi'],axis=1, inplace=True) #updating the same dataframe by dropping two columns


# In[9]:

cs_2009.reset_index(inplace=True)
cs_2010and2011.reset_index(inplace=True)
#cs_2011.reset_index(inplace=True)


# In[10]:

cs_2009['index']=cs_2009['index'].apply(lambda x:x.to_datetime())
cs_2009['year'] = cs_2009['index'].apply(lambda x:x.year)
cs_2009['month'] = cs_2009['index'].apply(lambda x:x.month)
cs_2009['day'] = cs_2009['index'].apply(lambda x:x.day)
cs_2009['hour'] = cs_2009['index'].apply(lambda x:x.hour)
cs_2009['min'] = cs_2009['index'].apply(lambda x:x.minute)


# In[11]:

cs_2010and2011['index']=cs_2010and2011['index'].apply(lambda x:x.to_datetime())
cs_2010and2011['year'] = cs_2010and2011['index'].apply(lambda x:x.year)
cs_2010and2011['month'] = cs_2010and2011['index'].apply(lambda x:x.month)
cs_2010and2011['day'] = cs_2010and2011['index'].apply(lambda x:x.day)
cs_2010and2011['hour'] = cs_2010and2011['index'].apply(lambda x:x.hour)
cs_2010and2011['min'] = cs_2010and2011['index'].apply(lambda x:x.minute)


# In[12]:

print(cs_2009.shape)
print(cs_2010and2011.shape)
#print(cs_2011.shape)


# In[13]:

cs_2009.drop(cs_2009.index[-1], inplace=True)
cs_2010and2011.drop(cs_2010and2011.index[-1], inplace=True)
#cs_2011.drop(cs_2011.index[-1], inplace=True)


# In[14]:

print(cs_2009.shape)
print(cs_2010and2011.shape)
#print(cs_2011.shape)


# In[15]:

cs_2010and2011.head()


# ### Import files from each year in a separate dataframe

# 
# - year            integer	 year, i.e., 1995
# - jday            integer	 Julian day (1 through 365 [or 366])
# - month           integer	 number of the month (1-12)
# - day             integer	 day of the month(1-31)
# - hour            integer	 hour of the day (0-23)
# - min             integer	 minute of the hour (0-59)
# - dt              real	 decimal time (hour.decimalminutes, e.g., 23.5 = 2330)
# - zen             real	 solar zenith angle (degrees)
# - dw_solar        real	 downwelling global solar (Watts m^-2)
# - uw_solar        real	 upwelling global solar (Watts m^-2)
# - direct_n        real	 direct-normal solar (Watts m^-2)
# - diffuse         real	 downwelling diffuse solar (Watts m^-2)
# - dw_ir           real	 downwelling thermal infrared (Watts m^-2)
# - dw_casetemp     real	 downwelling IR case temp. (K)
# - dw_dometemp     real	 downwelling IR dome temp. (K)
# - uw_ir           real	 upwelling thermal infrared (Watts m^-2)
# - uw_casetemp     real	 upwelling IR case temp. (K)
# - uw_dometemp     real	 upwelling IR dome temp. (K)
# - uvb             real	 global UVB (milliWatts m^-2)
# - par             real	 photosynthetically active radiation (Watts m^-2)
# - netsolar        real	 net solar (dw_solar - uw_solar) (Watts m^-2)
# - netir           real	 net infrared (dw_ir - uw_ir) (Watts m^-2)
# - totalnet        real	 net radiation (netsolar+netir) (Watts m^-2)
# - temp            real	 10-meter air temperature (?C)
# - rh              real	 relative humidity (%)
# - windspd         real	 wind speed (ms^-1)
# - winddir         real	 wind direction (degrees, clockwise from north)
# - pressure        real	 station pressure (mb)
# 

# In[16]:

cols = ['year', 'jday', 'month', 'day','hour','min','dt','zen','dw_solar','dw_solar_QC','uw_solar',
       'uw_solar_QC', 'direct_n','direct_n_QC','diffuse', 'diffuse_QC', 'dw_ir', 'dw_ir_QC', 'dw_casetemp',
       'dw_casetemp_QC', 'dw_dometemp','dw_dometemp_QC','uw_ir', 'uw_ir_QC', 'uw_casetemp','uw_casetemp_QC',
       'uw_dometemp','uw_dometemp_QC','uvb','uvb_QC','par','par_QC','netsolar','netsolar_QC','netir','netir_QC',
       'totalnet','totalnet_QC','temp','temp_QC','rh','rh_QC','windspd','windspd_QC','winddir','winddir_QC',
       'pressure','pressure_QC']


# In[17]:

path = r'./data/Desert_Rock/Exp_1_train'
all_files = glob.glob(path + "/*.dat")
all_files.sort()

df_big_train = pd.concat([pd.read_csv(f, skipinitialspace = True, quotechar = '"',skiprows=(2),delimiter=' ', 
                 index_col=False,header=None, names=cols) for f in all_files],ignore_index=True)
df_big_train.shape


# In[18]:

path = r'./data/Desert_Rock/Exp_1_test'
all_files = glob.glob(path + "/*.dat")
all_files.sort()

df_big_test = pd.concat((pd.read_csv(f, skipinitialspace = True, quotechar = '"',skiprows=(2),delimiter=' ', 
                 index_col=False,header=None, names=cols) for f in all_files),ignore_index=True)
df_big_test.shape


# In[19]:

df_big_test[df_big_test['dw_solar']==-9999.9].shape


# ### Merging Clear Sky GHI And the big dataframe

# In[20]:

df_train = pd.merge(df_big_train, cs_2010and2011, on=['year','month','day','hour','min'])
df_train.shape


# In[21]:

df_test = pd.merge(df_big_test, cs_2009, on=['year','month','day','hour','min'])
df_test.shape


# In[22]:

df_train.drop(['index'],axis=1, inplace=True) #updating the same dataframe by dropping the index columns from clear sky model
df_test.drop(['index'], axis=1, inplace=True)


# In[23]:

df_train.shape


# ### Managing missing values

# In[24]:

# Resetting index
df_train.reset_index(drop=True, inplace=True)
df_test.reset_index(drop=True, inplace=True)


# In[25]:

# Dropping rows with two or more -9999.9 values in columns


# In[26]:

# Step1: Get indices of all rows with 2 or more -999
missing_data_indices = np.where((df_train <=-9999.9).apply(sum, axis=1)>=2)[0]
# Step2: Drop those indices
df_train.drop(missing_data_indices, axis=0, inplace=True)
# Checking that the rows are dropped
df_train.shape


# In[27]:

missing_data_indices_test = np.where((df_test <= -9999.9).apply(sum, axis=1)>=2)[0]
df_test.drop(missing_data_indices_test, axis=0, inplace=True)
df_test.shape


# In[28]:

# For the rows with only one cell as -9999.9, replacing this cell with the mean of the column


# #### First resetting index after dropping rows in the previous part of the code

# In[29]:

# 2nd time - Reseting Index
df_train.reset_index(drop=True, inplace=True)
df_test.reset_index(drop=True, inplace=True)


# In[30]:

one_miss_train_idx = np.where((df_train <=-9999.9).apply(sum, axis=1)==1)[0]


# In[31]:

len(one_miss_train_idx)


# In[32]:

df_train.shape


# In[33]:

col_names = df_train.columns
from collections import defaultdict
stats = defaultdict(int)
total_single_missing_values = 0
for name in col_names:
    col_mean = df_train[~(df_train[name] == -9999.9)][name].mean()
    missing_indices = np.where((df_train[name] == -9999.9))
    stats[name] = len(missing_indices[0])
    df_train[name].loc[missing_indices] = col_mean
    total_single_missing_values += sum(df_train[name] == -9999.9)
    


# In[34]:

#df_col_min = df_train.apply(min, axis=0)
#df_col_max = df_train.apply(max, axis =0)
#print(df_col_min, df_col_max)


# In[35]:

train = np.where((df_train <=-9999.9).apply(sum, axis=1)==1)[0]


# In[36]:

len(train)


# In[37]:

# doing the same thing on test dataset


# In[38]:

one_miss_test_idx = np.where((df_test <=-9999.9).apply(sum, axis=1)==1)[0]
len(one_miss_test_idx)


# In[39]:

col_names_test = df_test.columns
from collections import defaultdict
stats_test = defaultdict(int)
total_single_missing_values_test = 0
for name in col_names_test:
    col_mean = df_test[~(df_test[name] == -9999.9)][name].mean()
    missing_indices = np.where((df_test[name] == -9999.9))
    stats_test[name] = len(missing_indices[0])
    df_test[name].loc[missing_indices] = col_mean
    total_single_missing_values_test += sum(df_test[name] == -9999.9)
    


# In[40]:

test = np.where((df_test <=-9999.9).apply(sum, axis=1)==1)[0]


# In[41]:

len(test)


# In[42]:

df_train.shape


# In[43]:

df_test.shape


# ### Exploratory Data Analysis

# In[ ]:

dw_solar_everyday = df_test.groupby(['jday'])['dw_solar'].mean()
ghi_everyday = df_test.groupby(['jday'])['ghi'].mean()
j_day = df_test['jday'].unique()


# In[ ]:

fig = plt.figure()

axes1 = fig.add_axes([0.1,0.1,0.8,0.8])
#axes2 = fig.add_axes([0.1,0.1,0.8,0.8])

axes1.scatter(j_day,dw_solar_everyday,label='Observed dw_solar',color='red')
axes1.scatter(j_day, ghi_everyday, label='Clear Sky GHI',color='green')

axes1.set_xlabel('Days')
axes1.set_ylabel('Solar Irradiance (Watts /m^2)')
axes1.set_title('Solar Irradiance - Test Year 2009')
axes1.legend(loc='best')

fig.savefig('./RNN Paper Results/Exp1_2/Desert_Rock/3hour_Figure 2.jpg', bbox_inches = 'tight')


# In[ ]:

sns.jointplot(x=dw_solar_everyday,y=ghi_everyday,kind='reg')
#plt.title('observed dw_solar vs clear sky ghi')
plt.xlabel('Observed global downwelling solar (Watts/m^2)')
plt.ylabel('Clear Sky GHI (Watts/m^2)')
plt.savefig('./RNN Paper Results/Exp1_2/Desert_Rock/3hour_Figure 3', bbox_inches='tight')


# ### making the Kt (clear sky index at time t) column by first removing rows with ghi==0

# In[44]:

df_train = df_train[df_train['ghi']!=0]
df_test = df_test[df_test['ghi']!=0]
df_train['Kt'] = df_train['dw_solar']/df_train['ghi']
df_test['Kt'] = df_test['dw_solar']/df_test['ghi']


# In[45]:

df_train.reset_index(inplace=True)
df_test.reset_index(inplace=True)


# In[46]:

print("test Kt max: "+str(df_test['Kt'].max()))
print("test Kt min: "+str(df_test['Kt'].min()))
print("test Kt mean: "+str(df_test['Kt'].mean()))
print("\n")
print("train Kt max: "+str(df_train['Kt'].max()))
print("train Kt min: "+str(df_train['Kt'].min()))
print("train Kt mean: "+str(df_train['Kt'].mean()))


# In[47]:

plt.plot(df_train['Kt'])


# In[48]:

plt.plot(df_test['Kt'])


# In[49]:

df_train= df_train[df_train['Kt']< 5000]
df_train= df_train[df_train['Kt']> -1000]
df_test= df_test[df_test['Kt']< 5000]
df_test= df_test[df_test['Kt']> -1000]


# #### Group the data (train dataframe)

# In[50]:

zen = df_train.groupby(['year','month','day','hour'])['zen'].mean()
dw_solar = df_train.groupby(['year','month','day','hour'])['dw_solar'].mean()
uw_solar = df_train.groupby(['year','month','day','hour'])['uw_solar'].mean()
direct_n = df_train.groupby(['year','month','day','hour'])['direct_n'].mean()
diffuse = df_train.groupby(['year','month','day','hour'])['diffuse'].mean()
dw_ir = df_train.groupby(['year','month','day','hour'])['dw_ir'].mean()
dw_casetemp = df_train.groupby(['year','month','day','hour'])['dw_casetemp'].mean()
dw_dometemp = df_train.groupby(['year','month','day','hour'])['dw_dometemp'].mean()
uw_ir = df_train.groupby(['year','month','day','hour'])['uw_ir'].mean()
uw_casetemp = df_train.groupby(['year','month','day','hour'])['uw_casetemp'].mean()
uw_dometemp = df_train.groupby(['year','month','day','hour'])['uw_dometemp'].mean()
uvb = df_train.groupby(['year','month','day','hour'])['uvb'].mean()
par = df_train.groupby(['year','month','day','hour'])['par'].mean()
netsolar = df_train.groupby(['year','month','day','hour'])['netsolar'].mean()
netir = df_train.groupby(['year','month','day','hour'])['netir'].mean()
totalnet = df_train.groupby(['year','month','day','hour'])['totalnet'].mean()
temp = df_train.groupby(['year','month','day','hour'])['temp'].mean()
rh = df_train.groupby(['year','month','day','hour'])['rh'].mean()
windspd = df_train.groupby(['year','month','day','hour'])['windspd'].mean()
winddir = df_train.groupby(['year','month','day','hour'])['winddir'].mean()
pressure = df_train.groupby(['year','month','day','hour'])['pressure'].mean()
ghi = df_train.groupby(['year','month','day','hour'])['ghi'].mean()
Kt = df_train.groupby(['year','month','day','hour'])['Kt'].mean()


# In[51]:

df_new_train = pd.concat([zen,dw_solar,uw_solar,direct_n,diffuse,dw_ir,dw_casetemp,dw_dometemp,uw_ir,uw_casetemp,uw_dometemp,
                    uvb,par,netsolar,netir,totalnet,temp,rh,windspd,winddir,pressure,ghi,Kt], axis=1)


# In[52]:

df_new_train.head()


# #### Groupdata - test dataframe

# In[53]:

test_zen = df_test.groupby(['month','day','hour'])['zen'].mean()
test_dw_solar = df_test.groupby(['month','day','hour'])['dw_solar'].mean()
test_uw_solar = df_test.groupby(['month','day','hour'])['uw_solar'].mean()
test_direct_n = df_test.groupby(['month','day','hour'])['direct_n'].mean()
test_diffuse = df_test.groupby(['month','day','hour'])['diffuse'].mean()
test_dw_ir = df_test.groupby(['month','day','hour'])['dw_ir'].mean()
test_dw_casetemp = df_test.groupby(['month','day','hour'])['dw_casetemp'].mean()
test_dw_dometemp = df_test.groupby(['month','day','hour'])['dw_dometemp'].mean()
test_uw_ir = df_test.groupby(['month','day','hour'])['uw_ir'].mean()
test_uw_casetemp = df_test.groupby(['month','day','hour'])['uw_casetemp'].mean()
test_uw_dometemp = df_test.groupby(['month','day','hour'])['uw_dometemp'].mean()
test_uvb = df_test.groupby(['month','day','hour'])['uvb'].mean()
test_par = df_test.groupby(['month','day','hour'])['par'].mean()
test_netsolar = df_test.groupby(['month','day','hour'])['netsolar'].mean()
test_netir = df_test.groupby(['month','day','hour'])['netir'].mean()
test_totalnet = df_test.groupby(['month','day','hour'])['totalnet'].mean()
test_temp = df_test.groupby(['month','day','hour'])['temp'].mean()
test_rh = df_test.groupby(['month','day','hour'])['rh'].mean()
test_windspd = df_test.groupby(['month','day','hour'])['windspd'].mean()
test_winddir = df_test.groupby(['month','day','hour'])['winddir'].mean()
test_pressure = df_test.groupby(['month','day','hour'])['pressure'].mean()
test_ghi = df_test.groupby(['month','day','hour'])['ghi'].mean()
test_Kt = df_test.groupby(['month','day','hour'])['Kt'].mean()


# In[54]:

df_new_test = pd.concat([test_zen,test_dw_solar,test_uw_solar,test_direct_n,test_diffuse,test_dw_ir,
                         test_dw_casetemp,test_dw_dometemp,test_uw_ir,test_uw_casetemp,test_uw_dometemp,
                    test_uvb,test_par,test_netsolar,test_netir,test_totalnet,test_temp,test_rh,
                         test_windspd,test_winddir,test_pressure,test_ghi,test_Kt], axis=1)


# In[55]:

df_new_test.loc[2].xs(17,level='day')


# ### Shifting Kt values to make 1 hour ahead forecast

# #### Train dataset

# In[56]:

levels_index= []
for m in df_new_train.index.levels:
    levels_index.append(m)


# In[57]:

for i in levels_index[0]:
    for j in levels_index[1]:
        df_new_train.loc[i].loc[j]['Kt'] = df_new_train.loc[i].loc[j]['Kt'].shift(-3)


# In[58]:

df_new_train = df_new_train[~(df_new_train['Kt'].isnull())]


# #### Test dataset

# In[59]:

levels_index2= []
for m in df_new_test.index.levels:
    levels_index2.append(m)


# In[60]:

for i in levels_index2[0]:
    for j in levels_index2[1]:
        df_new_test.loc[i].loc[j]['Kt'] = df_new_test.loc[i].loc[j]['Kt'].shift(-3)


# In[61]:

df_new_test = df_new_test[~(df_new_test['Kt'].isnull())]


# In[62]:

df_new_test[df_new_test['Kt']==df_new_test['Kt'].max()]


# ### Normalize train and test dataframe

# In[63]:

train_norm = (df_new_train - df_new_train.mean()) / (df_new_train.max() - df_new_train.min())
test_norm =  (df_new_test - df_new_test.mean()) / (df_new_test.max() - df_new_test.min())


# In[64]:

train_norm.reset_index(inplace=True,drop=True)
test_norm.reset_index(inplace=True,drop=True)


# ### Making train and test sets with train_norm and test_norm

# #### finding the gcf (greatest common factor) of train and test dataset's length and chop off the extra rows to make it divisible with the batchsize

# In[89]:

from fractions import gcd
gcd(train_norm.shape[0],test_norm.shape[0])


# In[65]:

import math
def roundup(x):
    return int(math.ceil(x / 100.0)) * 100


# In[66]:

train_lim = roundup(train_norm.shape[0])
test_lim = roundup(test_norm.shape[0])

train_random = train_norm.sample(train_lim-train_norm.shape[0])
test_random = test_norm.sample(test_lim-test_norm.shape[0])

train_norm = train_norm.append(train_random)
test_norm = test_norm.append(test_random)


# In[67]:

X1 = train_norm.drop('Kt',axis=1)
y1 = train_norm['Kt']

X2 = test_norm.drop('Kt',axis=1)
y2 = test_norm['Kt']


# In[68]:

print("X1_train shape is {}".format(X1.shape))
print("y1_train shape is {}".format(y1.shape))
print("X2_test shape is {}".format(X2.shape))
print("y2_test shape is {}".format(y2.shape))


# In[69]:

X_train = np.array(X1)
y_train  = np.array(y1)
X_test = np.array(X2)
y_test = np.array(y2)


# ### start of RNN

# In[70]:

import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision.datasets as dsets
from torch.autograd import Variable


# In[71]:

class RNNModel(nn.Module):
    def __init__(self, input_dim, hidden_dim, layer_dim, output_dim):
        super(RNNModel, self).__init__()
        #Hidden Dimension
        self.hidden_dim = hidden_dim
        
        # Number of hidden layers
        self.layer_dim = layer_dim
        
        #Building the RNN
        self.rnn = nn.RNN(input_dim, hidden_dim, layer_dim, batch_first=True, nonlinearity='relu')
        
        # Readout layer
        self.fc = nn.Linear(hidden_dim, output_dim)
        
    def forward(self, x):
        # Initializing the hidden state with zeros
        # (layer_dim, batch_size, hidden_dim)
        h0 = Variable(torch.zeros(self.layer_dim, x.size(0), self.hidden_dim))
        
        #One time step (the last one perhaps?)
        out, hn = self.rnn(x, h0)
        
        # Indexing hidden state of the last time step
        # out.size() --> ??
        #out[:,-1,:] --> is it going to be 100,100
        out = self.fc(out[:,-1,:])
        # out.size() --> 100,1
        return out
        


# In[72]:

# Instantiating Model Class
input_dim = 22
hidden_dim = 15
layer_dim = 1
output_dim = 1
batch_size = 100

model = RNNModel(input_dim, hidden_dim, layer_dim, output_dim)

# Instantiating Loss Class
criterion = nn.MSELoss()

# Instantiate Optimizer Class
learning_rate = 0.001
optimizer = torch.optim.SGD(model.parameters(), lr=learning_rate)

# converting numpy array to torch tensor
X_train = torch.from_numpy(X_train)
y_train = torch.from_numpy(y_train)
X_test = torch.from_numpy(X_test)
y_test = torch.from_numpy(y_test)

# initializing lists to store losses over epochs:
train_loss = []
test_loss = []
train_iter = []
test_iter = []


# In[73]:

# Training the model
seq_dim = 1

n_iter =0
num_samples = len(X_train)
test_samples = len(X_test)
batch_size = 100
num_epochs = 1000
feat_dim = X_train.shape[1]

X_train = X_train.type(torch.FloatTensor)
y_train = y_train.type(torch.FloatTensor)
X_test = X_test.type(torch.FloatTensor)
y_test = y_test.type(torch.FloatTensor)

for epoch in range(num_epochs):
    for i in range(0, int(num_samples/batch_size -1)):
        
        
        features = Variable(X_train[i*batch_size:(i+1)*batch_size, :]).view(-1, seq_dim, feat_dim)
        Kt_value = Variable(y_train[i*batch_size:(i+1)*batch_size])
        
        #print("Kt_value={}".format(Kt_value))
        
        optimizer.zero_grad()
        
        outputs = model(features)
        #print("outputs ={}".format(outputs))
        
        loss = criterion(outputs, Kt_value)
        
        train_loss.append(loss.data[0])
        train_iter.append(n_iter)

        #print("loss = {}".format(loss))
        loss.backward()
        
        optimizer.step()
        
        
            
        if n_iter%100 == 0:
            for i in range(0,int(test_samples/batch_size -1)):
                features = Variable(X_test[i*batch_size:(i+1)*batch_size, :]).view(-1, seq_dim, feat_dim)
                Kt_test = Variable(y_test[i*batch_size:(i+1)*batch_size])
                
                outputs = model(features)
                
                mse = np.sqrt(np.mean((Kt_test.data.numpy() - outputs.data.numpy().squeeze())**2)/num_samples)
                
                test_iter.append(n_iter)
                test_loss.append(mse)
                
            print('Epoch: {} Iteration: {}. Train_MSE: {}. Test_MSE: {}'.format(epoch, n_iter, loss.data[0], mse))
         
        n_iter += 1  


# In[74]:

print(len(test_loss))
#plt.plot(test_loss)
plt.plot(train_loss,'-')
#plt.ylim([0.000,0.99])


# In[75]:

plt.plot(test_loss,'r')


# #### Demornamization

# In[76]:

rmse = np.sqrt(mse)


# In[77]:

rmse_denorm = (rmse * (df_new_test['Kt'].max() - df_new_test['Kt'].min()))+ df_new_test['Kt'].mean()


# In[78]:

print("rmse_denorm",rmse_denorm)


# In[79]:

print(df_new_test['Kt'].describe())


# ### Saving train and test losses to a csv

# In[ ]:

df_trainLoss = pd.DataFrame(data={'Train Loss':train_loss,'iteration':train_iter}, columns=['Train Loss','iteration'])
df_trainLoss.to_csv('./RNN Paper Results/Exp1_2/Desert_Rock/3hour_DesertRock_TrainLoss.csv')
df_testLoss = pd.DataFrame(data={'Test Loss':test_loss,'iteration':test_iter}, columns=['Test Loss','iteration'])
df_testLoss.to_csv('./RNN Paper Results/Exp1_2/Desert_Rock/3hour_DesertRock_TestLoss.csv')