#!/usr/bin/env python3
# Copyright (c) Meta Platforms, Inc. and affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
from itertools import accumulate
from typing import Dict, List, Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
from torch.utils.data import DataLoader, Dataset, Subset
[docs]class CNN(nn.Module):
"""
Convolutional Neural Network.
"""
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(1, 20, kernel_size=5, stride=1)
self.fc1 = nn.Linear(8 * 8 * 20, 64)
self.fc2 = nn.Linear(64, 10)
[docs] def forward(self, x):
x = F.relu(self.conv1(x))
x = F.max_pool2d(x, 3, 3)
x = x.view(-1, 8 * 8 * 20)
x = F.relu(self.fc1(x))
x = self.fc2(x)
return F.log_softmax(x, dim=-1)
[docs]def load_mnist(
downsample_pct: float = 0.5,
train_pct: float = 0.8,
data_path: str = "./data",
batch_size: int = 128,
num_workers: int = 0,
deterministic_partitions: bool = False,
downsample_pct_test: Optional[float] = None,
) -> Tuple[DataLoader, DataLoader, DataLoader]:
"""
Load MNIST dataset (download if necessary) and split data into training,
validation, and test sets.
Args:
downsample_pct: the proportion of the dataset to use for training,
validation, and test
train_pct: the proportion of the downsampled data to use for training
data_path: Root directory of dataset where `MNIST/processed/training.pt`
and `MNIST/processed/test.pt` exist.
batch_size: how many samples per batch to load
num_workers: number of workers (subprocesses) for loading data
deterministic_partitions: whether to partition data in a deterministic
fashion
downsample_pct_test: the proportion of the dataset to use for test, default
to be equal to downsample_pct
Returns:
DataLoader: training data
DataLoader: validation data
DataLoader: test data
"""
# Specify transforms
transform = transforms.Compose(
[transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))]
)
# Load training set
train_valid_set = torchvision.datasets.MNIST(
root=data_path, train=True, download=True, transform=transform
)
# Load test set
test_set = torchvision.datasets.MNIST(
root=data_path, train=False, download=True, transform=transform
)
return get_partition_data_loaders(
train_valid_set=train_valid_set,
test_set=test_set,
downsample_pct=downsample_pct,
train_pct=train_pct,
batch_size=batch_size,
num_workers=num_workers,
deterministic_partitions=deterministic_partitions,
downsample_pct_test=downsample_pct_test,
)
[docs]def get_partition_data_loaders(
train_valid_set: Dataset,
test_set: Dataset,
downsample_pct: float = 0.5,
train_pct: float = 0.8,
batch_size: int = 128,
num_workers: int = 0,
deterministic_partitions: bool = False,
downsample_pct_test: Optional[float] = None,
) -> Tuple[DataLoader, DataLoader, DataLoader]:
"""
Helper function for partitioning training data into training and validation sets,
downsampling data, and initializing DataLoaders for each partition.
Args:
train_valid_set: torch.dataset
downsample_pct: the proportion of the dataset to use for training, and
validation
train_pct: the proportion of the downsampled data to use for training
batch_size: how many samples per batch to load
num_workers: number of workers (subprocesses) for loading data
deterministic_partitions: whether to partition data in a deterministic
fashion
downsample_pct_test: the proportion of the dataset to use for test, default
to be equal to downsample_pct
Returns:
DataLoader: training data
DataLoader: validation data
DataLoader: test data
"""
# Partition into training/validation
# pyre-ignore [6]
downsampled_num_examples = int(downsample_pct * len(train_valid_set))
n_train_examples = int(train_pct * downsampled_num_examples)
n_valid_examples = downsampled_num_examples - n_train_examples
train_set, valid_set, _ = split_dataset(
dataset=train_valid_set,
lengths=[
n_train_examples,
n_valid_examples,
len(train_valid_set) - downsampled_num_examples, # pyre-ignore [6]
],
deterministic_partitions=deterministic_partitions,
)
if downsample_pct_test is None:
downsample_pct_test = downsample_pct
# pyre-ignore [6]
downsampled_num_test_examples = int(downsample_pct_test * len(test_set))
test_set, _ = split_dataset(
test_set,
lengths=[
downsampled_num_test_examples,
len(test_set) - downsampled_num_test_examples, # pyre-ignore [6]
],
deterministic_partitions=deterministic_partitions,
)
train_loader = DataLoader(
train_set, batch_size=batch_size, shuffle=True, num_workers=num_workers
)
valid_loader = DataLoader(
valid_set, batch_size=batch_size, shuffle=False, num_workers=num_workers
)
test_loader = DataLoader(
test_set, batch_size=batch_size, shuffle=False, num_workers=num_workers
)
return train_loader, valid_loader, test_loader
[docs]def split_dataset(
dataset: Dataset, lengths: List[int], deterministic_partitions: bool = False
) -> List[Dataset]:
"""
Split a dataset either randomly or deterministically.
Args:
dataset: the dataset to split
lengths: the lengths of each partition
deterministic_partitions: deterministic_partitions: whether to partition
data in a deterministic fashion
Returns:
List[Dataset]: split datasets
"""
if deterministic_partitions:
indices = list(range(sum(lengths)))
else:
indices = torch.randperm(sum(lengths)).tolist()
return [
Subset(dataset, indices[offset - length : offset])
for offset, length in zip(accumulate(lengths), lengths)
]
[docs]def train(
net: torch.nn.Module,
train_loader: DataLoader,
parameters: Dict[str, float],
dtype: torch.dtype,
device: torch.device,
) -> nn.Module:
"""
Train CNN on provided data set.
Args:
net: initialized neural network
train_loader: DataLoader containing training set
parameters: dictionary containing parameters to be passed to the optimizer.
- lr: default (0.001)
- momentum: default (0.0)
- weight_decay: default (0.0)
- num_epochs: default (1)
dtype: torch dtype
device: torch device
Returns:
nn.Module: trained CNN.
"""
# Initialize network
net.to(dtype=dtype, device=device) # pyre-ignore [28]
net.train()
# Define loss and optimizer
criterion = nn.NLLLoss(reduction="sum")
optimizer = optim.SGD(
net.parameters(),
lr=parameters.get("lr", 0.001),
momentum=parameters.get("momentum", 0.0),
weight_decay=parameters.get("weight_decay", 0.0),
)
scheduler = optim.lr_scheduler.StepLR(
optimizer,
step_size=int(parameters.get("step_size", 30)),
gamma=parameters.get("gamma", 1.0), # default is no learning rate decay
)
num_epochs = parameters.get("num_epochs", 1)
# Train Network
# pyre-fixme[6]: Expected `int` for 1st param but got `float`.
for _ in range(num_epochs):
for inputs, labels in train_loader:
# move data to proper dtype and device
inputs = inputs.to(dtype=dtype, device=device)
labels = labels.to(device=device)
# zero the parameter gradients
optimizer.zero_grad()
# forward + backward + optimize
outputs = net(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
scheduler.step()
return net
[docs]def evaluate(
net: nn.Module, data_loader: DataLoader, dtype: torch.dtype, device: torch.device
) -> float:
"""
Compute classification accuracy on provided dataset.
Args:
net: trained model
data_loader: DataLoader containing the evaluation set
dtype: torch dtype
device: torch device
Returns:
float: classification accuracy
"""
net.eval()
correct = 0
total = 0
with torch.no_grad():
for inputs, labels in data_loader:
# move data to proper dtype and device
inputs = inputs.to(dtype=dtype, device=device)
labels = labels.to(device=device)
outputs = net(inputs)
_, predicted = torch.max(outputs.data, 1)
total += labels.size(0)
correct += (predicted == labels).sum().item()
return correct / total
