Pytorch Tutorial 3

simple linear regression with bulit in tools in pytorch

1. generate prediction
2. calculate the loss
3. compute gradients of w and b
4. adjust w and b
5. reset gradients to zero

these 5 steps also respect to the loop in the next function

import numpy as np
import torch.nn as nn
import torch
from torch.utils.data import TensorDataset
from torch.utils.data import DataLoader
import torch.nn.functional as F

# temp, rainfall, humidity
# inputs = torch.tensor(np.random.uniform(0, 120, size=(15, 3)))
# the input and output here need to specify the dtype, otherwise, when torch generate the prediction,
# it will encounter the problem of dtype is not match
inputs = torch.tensor(np.array(
    [[109.4144, 11.2775, 32.4521], [2.0002, 47.0248, 49.9469], [27.1528, 57.8907, 91.2076],
     [44.8227, 71.6239, 64.0752], [66.0968, 92.5966, 94.0775], [59.6257, 76.9701, 92.1656],
     [8.1551, 1.7426, 10.5297], [112.6036, 47.2793, 95.4221], [3.2212, 61.8274, 115.9187],
     [35.0351, 110.6133, 66.6992], [8.8387, 21.8008, 50.0480], [68.7698, 59.9815, 12.0230],
     [111.3881, 90.3050, 62.1327], [101.7462, 115.7447, 33.4925], [27.7659, 54.5803, 105.3599]], dtype='float32'))

# apples, oranges
# targets = torch.tensor(np.random.uniform(0, 50, size=(15, 2)))
targets = torch.tensor(np.array(
    [[28.1090, 45.0061], [29.0839, 6.4205], [35.2633, 44.1196],
     [29.5371, 6.8457], [7.4298, 36.1434], [6.6296, 47.1809],
     [49.9750, 49.9321], [34.1796, 16.6732], [46.8875, 7.6084],
     [23.0442, 42.2229], [29.7401, 13.4199], [3.0854, 21.4550],
     [47.6801, 49.1518], [18.7320, 18.4418], [34.2725, 25.8721]], dtype='float32'))
# print(inputs)
# print(targets)

# TensorDataset will creat the structure of pairing (input and target) accordingly
train_ds = TensorDataset(inputs, targets)

batch_size = 5
train_dl = DataLoader(train_ds, batch_size, shuffle=True)

# Each batch size is 5, and the data are shuffled
# and is still can contain the pair of data, the structure won't be shuffled
# for xb, yb in train_dl:
#     print("batch:")
#     print(xb)
#     print(yb)

# specify the input and output feature number
model = nn.Linear(3, 2)
# the weight and bias will be initialed automatically, and the parameter of requires_grad will be set as True
# print(model.weight)
# print(model.bias)
# print(list(model.parameters()))

# preds = model(inputs)
# print(preds)

loss_fn = F.mse_loss
loss = loss_fn(model(inputs), targets)
# print(loss)

opt = torch.optim.SGD(model.parameters(), lr=1e-5)

# 1 generate prediction
# 2 calculate the loss
# 3 compute gradients of w and b
# 4 adjust w and b
# 5 reset gradients to zero
# these 5 steps also respect to the loop in the next function

def fit(num_epochs, model, loss_fn, opt):
    # training interation
    for epoch in range(num_epochs):
        # batches in each interation
        for xb, yb in train_dl:
            pred = model(xb)
            loss = loss_fn(pred, yb)
            loss.backward()
            opt.step()
            opt.zero_grad()
        if (epoch+1) % 10 == 0:
            print('Epoch [{}/{}], Loss: {:.4f}'.format(epoch+1, num_epochs, loss.item()))

fit(100, model, loss_fn, opt)