The scorcher package provides high-level functionality for building
and fitting deep learning models using the Torch library in R. By
simplifying model development through easy-to-use functions and ensuring
compatibility with existing R workflows and data structures, scorcher
facilitates the use of deep learning without the need for extensive
programming knowledge.
The package allows users to create, modify, and visualize models through
functions such as initiate_scorch, scorch_layer, compile_scorch,
and fit_scorch. The package flexibly handles tasks such as prediction,
classification, computer vision, diffusion, and more.
You can install the development version of scorcher from GitHub with:
# install.packages("pak")
pak::pak("jtleek/scorcher")library(scorcher)Here is an example where we build a convolutional neural network using the MNIST dataset, which contains 70,000 grayscale images of handwritten digits, from 0 to 9. The goal is to classify these images into one of the 10 digit categories (0-9).
1. Prepare the Training Data:
library(torch)
library(torchvision)#- Training Data
train_data <- mnist_dataset(
root = tempdir(),
download = TRUE,
transform = transform_to_tensor)
x_train <- torch_tensor(train_data$data, dtype = torch_float()) |>
torch_unsqueeze(2)
y_train <- torch_tensor(train_data$targets, dtype = torch_long())Example Training Images:
Next, we’ll define our neural network using the scorcher package.
2. Create the Dataloader:
#- Create the Dataloader
dl <- scorch_create_dataloader(x_train, y_train, batch_size = 500)3. Define the Scorcher Model:
#- Define the Neural Network
scorch_model <- initiate_scorch(dl) |>
scorch_input("x") |>
scorch_layer("conv1", "conv2d", in_channels = 1, out_channels = 32, kernel_size = 3) |>
scorch_layer("act1", "relu") |>
scorch_layer("conv2", "conv2d", in_channels = 32, out_channels = 64, kernel_size = 3) |>
scorch_layer("act2", "relu") |>
scorch_layer("pool1", "max_pool2d", kernel_size = 2) |>
scorch_dropout("drop1", p = 0.25) |>
scorch_flatten("flat1") |>
scorch_layer("fc1", "linear", in_features = 9216, out_features = 128) |>
scorch_layer("act3", "relu") |>
scorch_layer("fc2", "linear", in_features = 128, out_features = 10) |>
scorch_output("fc2")Node names (like "conv1", "fc1") are unique identifiers that wire
the computation graph – they let nodes reference each other for
branching, fusion, and skip connections. See vignette("scorch_layer")
for naming conventions.
4. Compile the Model:
#- Compile the Neural Network
scorch_model <- scorch_model |>
compile_scorch(
loss_fn = nn_cross_entropy_loss(),
optimizer_fn = optim_adam,
optimizer_params = list(lr = 0.001)
)5. Train the Model
#-- Training the Neural Network
scorch_model <- scorch_model |>
fit_scorch(num_epochs = 10, verbose = TRUE)6. Evaluate the Model
Finally, we’ll evaluate our model on the testing data.
#- Testing Data
test_data <- mnist_dataset(
root = tempdir(),
train = FALSE,
transform = transform_to_tensor
)
x_test <- torch_tensor(test_data$data, dtype = torch_float()) |>
torch_unsqueeze(2)
y_test <- torch_tensor(test_data$targets, dtype = torch_long())
#- Model Predictions
scorch_model$nn_model$eval()
pred <- scorch_model$nn_model(x_test) |> torch_argmax(dim = 2)
accuracy <- sum(pred == y_test)$item() / length(y_test)
cat(sprintf("Testing Accuracy: %.2f%%\n", accuracy * 100))
#> Testing Accuracy: 98.59%Example Predictions Images:
Contributions are welcome! Please open an issue or submit a pull request on GitHub.
This project is licensed under the MIT License.