.gitignore

+/data
+*.pyc
+tests.ipynb
+

README.md

 # Image classification
 
 
-
-## Getting started
-
-To make it easy for you to get started with GitLab, here's a list of recommended next steps.
-
-Already a pro? Just edit this README.md and make it your own. Want to make it easy? [Use the template at the bottom](#editing-this-readme)!
-
-## Add your files
-
-- [ ] [Create](https://docs.gitlab.com/ee/user/project/repository/web_editor.html#create-a-file) or [upload](https://docs.gitlab.com/ee/user/project/repository/web_editor.html#upload-a-file) files
-- [ ] [Add files using the command line](https://docs.gitlab.com/ee/gitlab-basics/add-file.html#add-a-file-using-the-command-line) or push an existing Git repository with the following command:
-
-```
-cd existing_repo
-git remote add origin https://gitlab.ec-lyon.fr/barryt/image-classification.git
-git branch -M main
-git push -uf origin main
-```
-
-## Integrate with your tools
-
-- [ ] [Set up project integrations](https://gitlab.ec-lyon.fr/barryt/image-classification/-/settings/integrations)
-
-## Collaborate with your team
-
-- [ ] [Invite team members and collaborators](https://docs.gitlab.com/ee/user/project/members/)
-- [ ] [Create a new merge request](https://docs.gitlab.com/ee/user/project/merge_requests/creating_merge_requests.html)
-- [ ] [Automatically close issues from merge requests](https://docs.gitlab.com/ee/user/project/issues/managing_issues.html#closing-issues-automatically)
-- [ ] [Enable merge request approvals](https://docs.gitlab.com/ee/user/project/merge_requests/approvals/)
-- [ ] [Set auto-merge](https://docs.gitlab.com/ee/user/project/merge_requests/merge_when_pipeline_succeeds.html)
-
-## Test and Deploy
-
-Use the built-in continuous integration in GitLab.
-
-- [ ] [Get started with GitLab CI/CD](https://docs.gitlab.com/ee/ci/quick_start/index.html)
-- [ ] [Analyze your code for known vulnerabilities with Static Application Security Testing(SAST)](https://docs.gitlab.com/ee/user/application_security/sast/)
-- [ ] [Deploy to Kubernetes, Amazon EC2, or Amazon ECS using Auto Deploy](https://docs.gitlab.com/ee/topics/autodevops/requirements.html)
-- [ ] [Use pull-based deployments for improved Kubernetes management](https://docs.gitlab.com/ee/user/clusters/agent/)
-- [ ] [Set up protected environments](https://docs.gitlab.com/ee/ci/environments/protected_environments.html)
-
-***
-
-# Editing this README
-
-When you're ready to make this README your own, just edit this file and use the handy template below (or feel free to structure it however you want - this is just a starting point!). Thank you to [makeareadme.com](https://www.makeareadme.com/) for this template.
-
-## Suggestions for a good README
-Every project is different, so consider which of these sections apply to yours. The sections used in the template are suggestions for most open source projects. Also keep in mind that while a README can be too long and detailed, too long is better than too short. If you think your README is too long, consider utilizing another form of documentation rather than cutting out information.
-
-## Name
-Choose a self-explaining name for your project.
-
 ## Description
-Let people know what your project can do specifically. Provide context and add a link to any reference visitors might be unfamiliar with. A list of Features or a Background subsection can also be added here. If there are alternatives to your project, this is a good place to list differentiating factors.
-
-## Badges
-On some READMEs, you may see small images that convey metadata, such as whether or not all the tests are passing for the project. You can use Shields to add some to your README. Many services also have instructions for adding a badge.
-
-## Visuals
-Depending on what you are making, it can be a good idea to include screenshots or even a video (you'll frequently see GIFs rather than actual videos). Tools like ttygif can help, but check out Asciinema for a more sophisticated method.
-
-## Installation
-Within a particular ecosystem, there may be a common way of installing things, such as using Yarn, NuGet, or Homebrew. However, consider the possibility that whoever is reading your README is a novice and would like more guidance. Listing specific steps helps remove ambiguity and gets people to using your project as quickly as possible. If it only runs in a specific context like a particular programming language version or operating system or has dependencies that have to be installed manually, also add a Requirements subsection.
-
-## Usage
-Use examples liberally, and show the expected output if you can. It's helpful to have inline the smallest example of usage that you can demonstrate, while providing links to more sophisticated examples if they are too long to reasonably include in the README.
-
-## Support
-Tell people where they can go to for help. It can be any combination of an issue tracker, a chat room, an email address, etc.
-
-## Roadmap
-If you have ideas for releases in the future, it is a good idea to list them in the README.
-
-## Contributing
-State if you are open to contributions and what your requirements are for accepting them.
-
-For people who want to make changes to your project, it's helpful to have some documentation on how to get started. Perhaps there is a script that they should run or some environment variables that they need to set. Make these steps explicit. These instructions could also be useful to your future self.
-
-You can also document commands to lint the code or run tests. These steps help to ensure high code quality and reduce the likelihood that the changes inadvertently break something. Having instructions for running tests is especially helpful if it requires external setup, such as starting a Selenium server for testing in a browser.
+This BE is about image classification. The goal is to classify images from CIFAR10 dataset and evaluate the accuracy of the models, firstly by using a simple KNN and then by using a manually implemented MLP with one hidden layer. The MLP is implemented using numpy and the forward and backward passes are implemented using the chain rule. The MLP is trained using SGD and the gradients are computed using backpropagation. The MLP is trained using the CIFAR10 dataset and the accuracy is evaluated on the test set.
 
-## Authors and acknowledgment
-Show your appreciation to those who have contributed to the project.
+## Documentation
+The `be_image_classification.ipynb` file contains the main code used to train and evaluate the models.
 
-## License
-For open source projects, say how it is licensed.
+The functions implemented and used in this BE are located in the `utils` folder. 
 
-## Project status
-If you have run out of energy or time for your project, put a note at the top of the README saying that development has slowed down or stopped completely. Someone may choose to fork your project or volunteer to step in as a maintainer or owner, allowing your project to keep going. You can also make an explicit request for maintainers.
+The images located in the `images` folder are:
+- `knn_accuracy.png`: the evolution of the accuracy of the model with respect to the number of neighbors
+- `mlp_loss.png`: the evolution of the loss of the model with respect to the number of epochs, for the manually implemented MLP
+- `mlp_loss_tf.png`: the evolution of the accuracy of the equivalent model implemented using tensorflow with respect to the number of epochs to compare with the manually implemented MLP and check that the implementation is correct

utils/binary_cross_entropy.py

+import numpy as np
+
+def binary_cross_entropy(predictions: np.ndarray, targets: np.ndarray):
+    # compute the binary cross-entropy loss (1e-7 is added to avoid log(0))
+    predictions = np.clip(predictions, 1e-7, 1 - 1e-7)
+    term_0 = (1-targets) * np.log(1-predictions)
+    term_1 = targets * np.log(predictions)
+    return -np.mean(term_0+term_1)
\ No newline at end of file

utils/distance_matrix.py

+import numpy as np
+
+def distance_matrix(X: np.ndarray, Y: np.ndarray):
+    # compute the distance matrix between two sets of samples
+    x2 = np.sum(X**2, axis=1, keepdims=True)
+    y2 = np.sum(Y**2, axis=1, keepdims=True)
+    xy = X.dot(Y.T)
+    return np.sqrt(x2 - 2 * xy + y2.T)
\ No newline at end of file

utils/evaluate_knn.py

+from utils.knn_predict import knn_predict
+from utils.distance_matrix import distance_matrix
+import numpy as np
+
+def evaluate_knn(data_train, labels_train, data_test, labels_test, k):
+    # compute the accuracy of the kNN model
+    dists = distance_matrix(data_test, data_train)
+    y_pred = knn_predict(dists, labels_train, k)
+    accuracy = np.mean(y_pred == labels_test)
+    return accuracy
\ No newline at end of file

utils/forward_pass.py

+from utils.sigmoid import sigmoid
+import numpy as np
+
+def forward_pass(w1, b1, w2, b2, data):
+    # compute the forward pass of the MLP with sigmoid activations
+    z1 = np.matmul(data, w1) + b1
+    a1 = sigmoid(z1)
+    z2 = np.matmul(a1, w2) + b2
+    a2 = sigmoid(z2)
+    return a1, a2
\ No newline at end of file

utils/knn_predict.py

+import numpy as np
+
+def knn_predict(dists: np.ndarray, labels_train, k):
+    # predict labels based on k nearest neighbors
+    n = dists.shape[0]
+    y_pred = np.zeros(n, dtype=np.int64)
+    for i in range(n):
+        closest_y = []
+        closest_y = labels_train[np.argsort(dists[i])[:k]]
+        y_pred[i] = np.argmax(np.bincount(closest_y))
+    return y_pred
\ No newline at end of file

utils/learn_once_cross_entropy.py

+import numpy as np
+from utils.forward_pass import forward_pass
+from utils.binary_cross_entropy import binary_cross_entropy
+
+def adjust_weights_binary_cross_entropy(a1, a2, w1, b1, w2, b2, data, targets, learning_rate):
+    batch_size = data.shape[0]
+    dCdZ2 = a2 - targets
+    dCdW2 = np.matmul(a1.T, dCdZ2)  / batch_size
+    dCdB2 = np.sum(dCdZ2, axis=0, keepdims=True) / batch_size
+    dCdA1 = np.matmul(dCdZ2, w2.T)
+    dCdZ1 = dCdA1 * a1 * (1 - a1)
+    dCdW1 = np.matmul(data.T, dCdZ1) / batch_size
+    dCdB1 = np.sum(dCdZ1, axis=0, keepdims=True) / batch_size
+
+    w2 -= learning_rate * dCdW2
+    w1 -= learning_rate * dCdW1
+    b1 -= learning_rate * dCdB1
+    b2 -= learning_rate * dCdB2
+    return w1, b1, w2, b2
+
+def learn_once_cross_entropy(w1,b1,w2,b2,data,targets,learning_rate):
+    a1, a2 = forward_pass(w1, b1, w2, b2, data)
+    loss = binary_cross_entropy(a2, targets)
+    w1, b1, w2, b2 = adjust_weights_binary_cross_entropy(a1, a2, w1, b1, w2, b2, data, targets, learning_rate)
+    return w1, b1, w2, b2, loss
\ No newline at end of file

utils/learn_once_mse.py

+import numpy as np
+from utils.forward_pass import forward_pass
+
+def adjust_weights_mse(a1, a2, w1, b1, w2, b2, data, targets, learning_rate):
+    batch_size = data.shape[0]
+    N_out = targets.shape[1]
+    dCdA2 = 2 * (a2 - targets) / N_out
+    dCdZ2 = dCdA2 * a2 * (1 - a2)
+    dCdW2 = np.matmul(a1.T, dCdZ2) / batch_size 
+    dCdB2 = np.sum(dCdZ2, axis=0, keepdims=True) / batch_size
+    dCdA1 = np.matmul(dCdZ2, w2.T) 
+    dCdZ1 = dCdA1 * a1 * (1 - a1)
+    dCdW1 = np.matmul(data.T, dCdZ1) / batch_size
+    dCdB1 = np.sum(dCdZ1, axis=0, keepdims=True) / batch_size
+
+    w2 -= learning_rate * dCdW2
+    w1 -= learning_rate * dCdW1
+    b1 -= learning_rate * dCdB1
+    b2 -= learning_rate * dCdB2
+    return w1, b1, w2, b2
+
+
+def learn_once_mse(w1: np.ndarray, b1: np.ndarray, w2: np.ndarray, b2: np.ndarray, data: np.ndarray, targets: np.ndarray, learning_rate: float):
+    a1, a2 = forward_pass(w1, b1, w2, b2, data)
+    loss = np.mean(np.square(a2 - targets))
+    w1, b1, w2, b2 = adjust_weights_mse(a1, a2, w1, b1, w2, b2, data, targets, learning_rate)
+    return w1, b1, w2, b2, loss
\ No newline at end of file

utils/mlp_training.py

+import tqdm
+import numpy as np
+from utils.forward_pass import forward_pass
+from utils.one_hot import one_hot
+from utils.learn_once_cross_entropy import learn_once_cross_entropy
+
+
+def train_mlp(w1, b1, w2, b2, data_train, labels_train, learning_rate, num_epochs, batch_size, n_classes):
+    # train the MLP for num_epochs epochs, using batches of size batch_size
+    losses = []
+    for epoch in range(num_epochs):
+        for i in tqdm.tqdm(range(0, data_train.shape[0], batch_size)):
+            data = data_train[i:i+batch_size]
+            targets = one_hot(labels_train[i:i+batch_size], n_classes)
+            w1, b1, w2, b2, loss = learn_once_cross_entropy(w1, b1, w2, b2, data, targets, learning_rate)
+        losses.append(loss)
+        print(f'epoch={epoch}, loss={loss}')
+    return losses, w1, b1, w2, b2
+
+def test_mlp(w1, b1, w2, b2, data_test, labels_test):
+    # test the MLP on data_test, and return the accuracy
+    _, a2 = forward_pass(w1, b1, w2, b2, data_test)
+    predictions = np.argmax(a2, axis=1)
+    test_accuracy = np.mean(predictions == labels_test)
+    return test_accuracy
+
+def initialize_mlp(d_in, d_h, d_out):
+    # initialize the weights and biases of the MLP
+    w1 = 2 * np.random.rand(d_in, d_h) - 1
+    b1 = np.zeros((1, d_h))
+    w2 = 2 * np.random.rand(d_h, d_out) - 1
+    b2 = np.zeros((1, d_out))
+    return w1, b1, w2, b2
+
+def run_mlp_training(data_train, labels_train, data_test, labels_test, d_h, learning_rate, num_epochs, batch_size, n_classes = 10):
+    # run the training and testing of the MLP on data_train and data_test
+    d_in = data_train.shape[1]
+    d_out = np.max(labels_train) + 1
+    w1, b1, w2, b2 = initialize_mlp(d_in, d_h, d_out)
+    losses, w1, b1, w2, b2 = train_mlp(w1, b1, w2, b2, data_train, labels_train, learning_rate, num_epochs, batch_size, n_classes)
+    test_accuracy = test_mlp(w1, b1, w2, b2, data_test, labels_test)
+    return losses, test_accuracy
\ No newline at end of file

utils/one_hot.py

+import numpy as np
+
+def one_hot(labels: np.ndarray, n_classes: int):
+    # convert an array of labels to a one-hot representation
+    n = labels.shape[0]
+    one_hot_labels = np.zeros((n, n_classes))
+    one_hot_labels[np.arange(n), labels] = 1
+    return one_hot_labels
\ No newline at end of file

utils/process_image.py

+import matplotlib.pyplot as plt
+def plot_image_with_label(img, label):
+    # plot image with label
+    plt.imshow(img)
+    plt.title(label)
+    plt.show()
+
+def save_plot_as_image(X, Y, y_label, x_label, save_path):
+    # plot and save image as png
+    plt.figure(figsize=(10,5))
+    plt.plot(X, Y)
+    plt.ylabel(y_label)
+    plt.xlabel(x_label)
+    plt.savefig(save_path)
+    plt.show()
+    plt.close()
+    
\ No newline at end of file

utils/read_cifar.py

+import glob
+import numpy as np
+import pickle
+
+def read_cifar_batch(batch_path):
+    # read a batch of cifar data
+    with open(batch_path, 'rb') as f:
+        batch = pickle.load(f, encoding='bytes')
+    data=np.array(batch[b'data'],dtype=np.float32)/255.0
+    labels=np.array(batch[b'labels'],dtype=np.int64)
+  
+    return data, labels
+
+def read_cifar(directory):
+    # read all cifar data in a directory
+    files = glob.glob(f'{directory}/*_batch*')
+    data = np.empty((0, 3072), dtype=np.float32)
+    labels = np.empty((0), dtype=np.int64)
+    for file in files:
+        batch_data, batch_labels = read_cifar_batch(file)
+        data = np.vstack((data, batch_data))
+        labels = np.hstack((labels, batch_labels))
+    #print(data.shape, labels.shape)
+    return data, labels

utils/sigmoid.py

+import numpy as np
+def sigmoid(x):
+    return 1 / (1 + np.exp(-x))
\ No newline at end of file

utils/split_dataset.py

+import numpy as np
+
+def split_dataset(data, labels, split: float)-> (np.ndarray, np.ndarray, np.ndarray, np.ndarray):
+    # split dataset into train and test
+    assert data.shape[0] == labels.shape[0]
+    # shuffle data
+    indices = np.arange(data.shape[0])
+    np.random.shuffle(indices)
+    data = data[indices]
+    labels = labels[indices]
+    # split data
+    split_index = int(data.shape[0] * split)
+    train_data = data[:split_index]
+    train_labels = labels[:split_index]
+    test_data = data[split_index:]
+    test_labels = labels[split_index:]
+    return train_data, train_labels, test_data, test_labels
\ No newline at end of file