Challenge dataset

README.md

@@ -1,3 +1,50 @@
----
-license: mit
----
+---
+license: mit
+---
+
+# MNIST dataset used during the Perceval Quest challenge
+
+This repository hosts a partial MNIST dataset used during the Perceval Quest as part of the
+Hybrid AI Quantum Challenge. The dataset is stored under `data/` and split into
+`train.csv` and `val.csv`.
+
+This dataset is a subset of the original MNIST dataset that can be found [here](https://web.archive.org/web/20200430193701/http://yann.lecun.com/exdb/mnist/) and introduced in [LeCun et al., 1998a].
+
+The Perceval Quest challenge lasted from November 2024 to March 2025. More than 64 teams participated in its first phase and 12 teams were selected amongst the finalist.
+
+## Dataset structure
+
+- `data/train.csv`
+- `data/val.csv`
+
+Each CSV contains two columns:
+
+- `image`: a stringified list of 784 floats (28x28 grayscale image)
+- `label`: the digit class (0-9)
+
+## Load the dataset from `data/`
+
+### Option 1: pandas
+
+```python
+import pandas as pd
+
+train_df = pd.read_csv("./data/train.csv")
+val_df = pd.read_csv("./data/val.csv")
+```
+
+### Option 2: PyTorch Dataset (provided)
+
+```python
+from data_utils import MNIST_partial
+
+train_set = MNIST_partial(data="./data", split="train")
+val_set = MNIST_partial(data="./data", split="val")
+```
+
+## References
+- Dataset: [LeCun et al., 1998a] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. "Gradient-based learning applied to document recognition." Proceedings of the IEEE, 86(11):2278-2324, November 1998.
+- Paper: NOTTON, Cassandre, APOSTOLOU, Vassilis, SENELLART, Agathe, et al.
+  Establishing Baselines for Photonic Quantum Machine Learning: Insights from an Open,
+  Collaborative Initiative. arXiv preprint arXiv:2510.25839, 2025.
+- Repository: https://github.com/Quandela/HybridAIQuantum-Challenge/tree/main

data/train.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a0e126569c49d8c7130f5695d33a7d67f83c19ba55458233aabb6141ef306c0
+size 29292986

data_utils.py

@@ -0,0 +1,90 @@
+from torch.utils.data import Dataset, DataLoader
+import os
+import pandas as pd
+import re
+import torch
+import torchvision.transforms as transforms
+import matplotlib.pyplot as plt
+
+################
+## DATA UTILS ##
+################
+
+# load the correct train, val dataset for the challenge, from the csv files
+class MNIST_partial(Dataset):
+    def __init__(self, data = './data', transform=None, split = 'train'):
+        """
+        Args:
+            data: path to dataset folder which contains train.csv and val.csv
+            transform (callable, optional): Optional transform to be applied
+                on a sample (e.g., data augmentation or normalization)
+            split: 'train' or 'val' to determine which set to download
+        """
+        self.data_dir = data
+        self.transform = transform
+        self.data = []
+        
+        if split == 'train':
+            filename = os.path.join(self.data_dir,'train.csv')
+        elif split == 'val':
+            filename = os.path.join(self.data_dir,'val.csv')
+        else:
+            raise AttributeError("split!='train' and split!='val': split must be train or val")
+        
+        self.df = pd.read_csv(filename)
+        
+    
+    def __len__(self):
+        l = len(self.df['image'])
+        return l
+    
+    def __getitem__(self, idx):
+        img = self.df['image'].iloc[idx]
+        label = self.df['label'].iloc[idx]
+        # string to list
+        img_list = re.split(r',', img)
+        # remove '[' and ']'
+        img_list[0] = img_list[0][1:]
+        img_list[-1] = img_list[-1][:-1]
+        # convert to float
+        img_float = [float(el) for el in img_list]
+        # convert to image
+        img_square = torch.unflatten(torch.tensor(img_float),0,(1,28,28))
+        if self.transform is not None:
+            img_square = self.transform(img_square)
+        return img_square, label
+
+
+
+####################
+## TRAINING UTILS ##
+####################
+
+# plot the training curves (accuracy and loss) and save them in 'training_curves.png'
+def plot_training_metrics(train_acc,val_acc,train_loss,val_loss):
+    fig, axes = plt.subplots(1,2,figsize = (15,5))
+    X = [i for i in range(len(train_acc))]
+    names = [str(i+1) for i in range(len(train_acc))]
+    axes[0].plot(X,train_acc,label = 'training')
+    axes[0].plot(X,val_acc,label = 'validation')
+    axes[0].set_xlabel("Epochs")
+    axes[0].set_ylabel("ACC")
+    axes[0].set_title("Training and validation accuracies")
+    axes[0].grid(visible = True)
+    axes[0].legend()
+    axes[1].plot(X,train_loss,label = 'training')
+    axes[1].plot(X,val_loss,label = 'validation')
+    axes[1].set_xlabel("Epochs")
+    axes[1].set_ylabel("Loss")
+    axes[1].set_title("Training and validation losses")
+    axes[1].grid(visible = True)
+    axes[1].legend()
+    axes[0].set_xticks(ticks=X,labels = names)
+    axes[1].set_xticks(ticks=X,labels = names)
+    fig.savefig("training_curves.png")
+
+
+# compute the accuracy of the model
+def accuracy(outputs, labels):
+    _, preds = torch.max(outputs, dim = 1)
+    return(torch.tensor(torch.sum(preds == labels).item()/ len(preds)))