Pretrained Image classification example - Transformer¶

• Pretrained image classification using a Transformer architecture, "custom" Optimizer for OpenML Task (362128) , tiniest ImageNet dataset.

In [ ]:

import torch.nn
import torch.optim

import openml
import openml_pytorch
import openml_pytorch.layers
import openml_pytorch.config
from openml import OpenMLTask
import logging
import warnings
from torchvision.transforms import Compose, Resize, ToPILImage, ToTensor, Lambda
from openml_pytorch.trainer import convert_to_rgb
# Suppress FutureWarning messages
warnings.simplefilter(action='ignore')

############################################################################
# Enable logging in order to observe the progress while running the example.
openml.config.logger.setLevel(logging.DEBUG)
openml_pytorch.config.logger.setLevel(logging.DEBUG)
############################################################################

############################################################################
import torch.nn as nn
import torch.nn.functional as F

import torch.nn import torch.optim import openml import openml_pytorch import openml_pytorch.layers import openml_pytorch.config from openml import OpenMLTask import logging import warnings from torchvision.transforms import Compose, Resize, ToPILImage, ToTensor, Lambda from openml_pytorch.trainer import convert_to_rgb # Suppress FutureWarning messages warnings.simplefilter(action='ignore') ############################################################################ # Enable logging in order to observe the progress while running the example. openml.config.logger.setLevel(logging.DEBUG) openml_pytorch.config.logger.setLevel(logging.DEBUG) ############################################################################ ############################################################################ import torch.nn as nn import torch.nn.functional as F

In [ ]:

# openml.config.apikey = 'key'
from openml_pytorch.trainer import OpenMLTrainerModule
from openml_pytorch.trainer import OpenMLDataModule
from openml_pytorch.trainer import Callback

# openml.config.apikey = 'key' from openml_pytorch.trainer import OpenMLTrainerModule from openml_pytorch.trainer import OpenMLDataModule from openml_pytorch.trainer import Callback

Define the Model¶

In [ ]:

# Example model. You can do better :)
import torchvision.models as models

# Load the pre-trained ResNet model
model = models.efficientnet_b0(pretrained=True)

# Modify the last fully connected layer to the required number of classes
num_classes = 200
in_features = model.classifier[-1].in_features
# model.fc = nn.Linear(in_features, num_classes)
model.classifier = nn.Sequential(
    nn.Dropout(p=0.2, inplace=True),
    nn.Linear(in_features, num_classes),
)

# Optional: If you're fine-tuning, you may want to freeze the pre-trained layers
# for param in model.parameters():
#     param.requires_grad = False

# # If you want to train the last layer only (the newly added layer)
# for param in model.fc.parameters():
#     param.requires_grad = True

# Example model. You can do better :) import torchvision.models as models # Load the pre-trained ResNet model model = models.efficientnet_b0(pretrained=True) # Modify the last fully connected layer to the required number of classes num_classes = 200 in_features = model.classifier[-1].in_features # model.fc = nn.Linear(in_features, num_classes) model.classifier = nn.Sequential( nn.Dropout(p=0.2, inplace=True), nn.Linear(in_features, num_classes), ) # Optional: If you're fine-tuning, you may want to freeze the pre-trained layers # for param in model.parameters(): # param.requires_grad = False # # If you want to train the last layer only (the newly added layer) # for param in model.fc.parameters(): # param.requires_grad = True

Configure the Data Module¶

• Make sure the data is present in the file_dir directory, and the filename_col is correctly set along with this column correctly pointing to where your data is stored.

In [ ]:

transform = Compose(
    [
        ToPILImage(),  # Convert tensor to PIL Image to ensure PIL Image operations can be applied.
        Lambda(
            convert_to_rgb
        ),  # Convert PIL Image to RGB if it's not already.
        Resize(
            (64, 64)
        ),  # Resize the image.
        ToTensor(),  # Convert the PIL Image back to a tensor.
    ]
)
data_module = OpenMLDataModule(
    type_of_data="image",
    file_dir="datasets",
    filename_col="image_path",
    target_mode="categorical",
    target_column="label",
    batch_size = 64,
    transform=transform
)

transform = Compose( [ ToPILImage(), # Convert tensor to PIL Image to ensure PIL Image operations can be applied. Lambda( convert_to_rgb ), # Convert PIL Image to RGB if it's not already. Resize( (64, 64) ), # Resize the image. ToTensor(), # Convert the PIL Image back to a tensor. ] ) data_module = OpenMLDataModule( type_of_data="image", file_dir="datasets", filename_col="image_path", target_mode="categorical", target_column="label", batch_size = 64, transform=transform )

Configure the Trainer Module¶

In [ ]:

def custom_optimizer_gen(model: torch.nn.Module, task: OpenMLTask) -> torch.optim.Optimizer:
    return torch.optim.Adam(model.fc.parameters())

trainer = OpenMLTrainerModule(
    data_module=data_module,
    verbose = True,
    epoch_count = 1,
    optimizer = custom_optimizer_gen,
    callbacks=[],
)
openml_pytorch.config.trainer = trainer

def custom_optimizer_gen(model: torch.nn.Module, task: OpenMLTask) -> torch.optim.Optimizer: return torch.optim.Adam(model.fc.parameters()) trainer = OpenMLTrainerModule( data_module=data_module, verbose = True, epoch_count = 1, optimizer = custom_optimizer_gen, callbacks=[], ) openml_pytorch.config.trainer = trainer

Download the task¶

In [ ]:

# Download the OpenML task for tiniest imagenet
task = openml.tasks.get_task(362128)

# Download the OpenML task for tiniest imagenet task = openml.tasks.get_task(362128)

Run the model on the task¶

In [ ]:

#
# Run the model on the task (requires an API key).m
run = openml.runs.run_model_on_task(model, task, avoid_duplicate_runs=False)

# # Run the model on the task (requires an API key).m run = openml.runs.run_model_on_task(model, task, avoid_duplicate_runs=False)

View loss¶

In [ ]:

trainer.runner.cbs[1].plot_loss()

trainer.runner.cbs[1].plot_loss()

View learning rate¶

In [ ]:

trainer.runner.cbs[1].plot_lr()

trainer.runner.cbs[1].plot_lr()

Publish the run to OpenML¶

In [ ]:

run.publish()

run.publish()