LSTM-a1_temperatura.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Sat Sep 28 22:57:42 2019

@author: jeremiah
"""
#%%
import torch as t
from torch import nn
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline


#%% reading the data

df = pd.read_csv('/home/jeremiah/data_analysis/datasets/time-series-pam/a1_temperatura.csv', index_col = 0)

df.head()
uba_data = np.zeros(len(df['Ubatuba']))
uba_data = df['Ubatuba'].values

#%%

train_data, val_data = uba_data[:100], uba_data[100:]


x_train, y_train = train_data[:-1], train_data[1:]

x_val, y_val = val_data[:-1], val_data[1:]

plt.plot(x_train, 'r')
plt.plot(y_train, 'b')


#%%

class RNN(nn.Module):
    def __init__(self, input_size, hidden_size, n_layers, dropout_p = 0.5):
        super().__init__()
        self.hidden_size = hidden_size

        self.lstm = nn.LSTM(input_size, self.hidden_size, n_layers, batch_first = True)

        self.fc = nn.Linear(self.hidden_size, 1)

        self.dropout = nn.Dropout(dropout_p)

    def forward(self, x, h):

        out, h = self.lstm(x, h)

        out = self.dropout(out)

        out = out.view(-1, self.hidden_size)

        out = self.fc(out)

        return out, h

#%%
input_size = 1
hidden_size = 16
n_layers = 1

net = RNN(input_size, hidden_size, n_layers)
print(net)

#%%
from torch import optim

criterion = nn.MSELoss()
op = optim.Adam(net.parameters(), lr=0.001)

#%%

def train(net, steps, print_every, x, y):
    weight = next(net.parameters()).data
    hidden = (weight.new(1, 1, 16).zero_(), weight.new(1, 1, 16).zero_())
    for s in range(steps):
        x = t.Tensor(x)
        x = x.view(1, 99, 1)
        y = t.Tensor(y)

        hidden = tuple([each.data for each in hidden])
        print(x.shape)
        out, hidden = net(x, hidden)
        y = y.view(*out.shape)
        print(out.shape)
        print(y.shape)

        loss = criterion(out, y)
        op.zero_grad()
        loss.backward()
        op.step()

        if s % print_every == 0:

            plt.plot(x.view(-1), 'r')
            plt.plot(out.data, 'g')
            plt.plot(y, 'b')
            plt.show()

    return net

#%%
steps = 5
print_every = 1


net = train(net, steps, print_every, x_train, y_train)