feat: svm model

evaluate.py

@@ -0,0 +1,69 @@
+import argparse
+import csv
+
+from sklearn.metrics import accuracy_score, confusion_matrix
+
+parser = argparse.ArgumentParser()
+
+parser.add_argument(
+    "--ground-truth",
+    required=True
+)
+
+parser.add_argument(
+    "--predictions",
+    required=True
+)
+
+args = parser.parse_args()
+
+ground_truth_csv = args.ground_truth
+predictions_csv = args.predictions
+
+gt = {}
+
+with open(ground_truth_csv, newline="") as f:
+    reader = csv.reader(f)
+    header = next(reader)
+
+    for row in reader:
+        path = row[0]
+        label = row[1]
+        gt[path] = label
+
+
+preds = {}
+
+with open(predictions_csv, newline="") as f:
+    reader = csv.reader(f)
+    header = next(reader)
+
+    for row in reader:
+        path = row[0]
+        label = row[1]
+        preds[path] = label
+
+
+y_true = []
+y_pred = []
+
+for path, true_label in gt.items():
+    if path in preds:
+        y_true.append(true_label)
+        y_pred.append(preds[path])
+
+
+acc = accuracy_score(y_true, y_pred)
+
+cm = confusion_matrix(y_true, y_pred)
+
+print("Accuracy:", acc)
+print("Confusion matrix:")
+for row in cm:
+    print(" ".join(str(x) for x in row))
+
+
+
+
+
+

mnist.py

@@ -0,0 +1,98 @@
+import csv
+import argparse
+
+import numpy as np
+from PIL import Image
+from sklearn.svm import SVC
+from joblib import dump, load
+
+
+def load_image_as_vector(path):
+    img = Image.open(path).convert('L')
+    arr = np.array(img)
+    return arr.flatten()
+
+
+
+parser = argparse.ArgumentParser()
+
+parser.add_argument("--mode", required=True)
+parser.add_argument("--dataset")
+parser.add_argument("--model", required=True)
+parser.add_argument("--input")
+parser.add_argument("--output")
+
+
+args = parser.parse_args()
+
+if args.mode == "train":
+    dataset_csv = args.dataset
+    model_path = args.model
+
+    paths = []
+    labels = []
+
+    with open(dataset_csv, newline="") as f:
+        reader = csv.reader(f)
+        header = next(reader)
+
+        for row in reader:
+            path = row[0]
+            label = row[1]
+            paths.append(path)
+            labels.append(label)
+
+
+    X_list = []
+    Y_list = []
+
+
+    for path, label in zip(paths, labels):
+        vec = load_image_as_vector(path)
+        X_list.append(vec)
+        Y_list.append(label)
+
+    X = np.array(X_list)
+    Y = np.array(Y_list)
+
+
+    model = SVC(kernel="rbf", gamma="scale")
+    model.fit(X, Y)
+
+
+    dump(model, model_path)
+
+elif args.mode == "inference":
+    model_path = args.model
+    input_csv = args.input
+    output_csv = args.output
+
+    model = load(model_path)
+
+    paths = []
+
+    with open(input_csv, newline="") as f:
+        reader = csv.reader(f)
+        header = next(reader)
+
+        for row in reader:
+            path = row[0]
+            paths.append(path)
+
+    X_list = []
+
+    for path in paths:
+        vec = load_image_as_vector(path)
+        X_list.append(vec)
+
+    X = np.array(X_list)
+
+    preds = model.predict(X)
+
+    with open(output_csv, 'w', newline="") as f:
+        writer = csv.writer(f)
+        writer.writerow(['path', 'label'])
+        for path, label in zip(paths, preds):
+            writer.writerow([path, label])
+
+

prepare_from_hf.py

@@ -0,0 +1,99 @@
+from datasets import load_dataset
+import os
+import csv
+
+
+mnist = load_dataset("ylecun/mnist")
+mnist_train = mnist["train"]
+mnist_test = mnist["test"]
+
+
+MNIST_TRAIN_DIR = "mnist_images_train"
+MNIST_TEST_DIR = "mnist_images_test"
+
+MNIST_TRAIN_CSV = "mnist_train.csv"
+MNIST_TEST_CSV = "mnist_test.csv"
+
+os.makedirs(MNIST_TRAIN_DIR, exist_ok=True)
+os.makedirs(MNIST_TEST_DIR, exist_ok=True)
+
+with open(MNIST_TRAIN_CSV, "w", newline="") as f:
+    writer = csv.writer(f)
+    writer.writerow(["path", "label"])
+
+    for idx, item in enumerate(mnist_train):
+        img = item["image"]
+        label = item["label"]
+
+        filename = f"mnist_train_{idx:05d}_{label}.png"
+        img_path = os.path.join(MNIST_TRAIN_DIR, filename)
+
+        img.save(img_path)
+        writer.writerow([img_path, label])
+
+
+with open(MNIST_TEST_CSV, "w", newline="") as f:
+    writer = csv.writer(f)
+    writer.writerow(["path", "label"])
+
+    for idx, item in enumerate(mnist_test):
+        img = item["image"]
+        label = item["label"]
+
+        filename = f"mnist_test_{idx:05d}_{label}.png"
+        img_path = os.path.join(MNIST_TEST_DIR, filename)
+
+        img.save(img_path)
+        writer.writerow([img_path, label])
+
+
+fashion = load_dataset("fashion_mnist")
+fashion_train = fashion["train"]
+fashion_test = fashion["test"]
+
+FASHION_TRAIN_DIR = "fashion_images_train"
+FASHION_TEST_DIR = "fashion_images_test"
+
+FASHION_TRAIN_CSV = "fashion_train.csv"
+FASHION_TEST_CSV = "fashion_test.csv"
+
+os.makedirs(FASHION_TRAIN_DIR, exist_ok=True)
+os.makedirs(FASHION_TEST_DIR, exist_ok=True)
+
+with open(FASHION_TRAIN_CSV, "w", newline="") as f:
+    writer = csv.writer(f)
+    writer.writerow(["path", "label"])
+
+    for idx, item in enumerate(fashion_train):
+        img = item["image"]
+        label = item["label"]
+
+        filename = f"fashion_train_{idx:05d}_{label}.png"
+        img_path = os.path.join(FASHION_TRAIN_DIR, filename)
+
+        img.save(img_path)
+        writer.writerow([img_path, label])
+
+
+with open(FASHION_TEST_CSV, "w", newline="") as f:
+    writer = csv.writer(f)
+    writer.writerow(["path", "label"])
+
+    for idx, item in enumerate(fashion_test):
+        img = item["image"]
+        label = item["label"]
+
+        filename = f"fashion_test_{idx:05d}_{label}.png"
+        img_path = os.path.join(FASHION_TEST_DIR, filename)
+
+        img.save(img_path)
+        writer.writerow([img_path, label])
+
+
+
+
+
+
+
+
+

requirements.txt

@@ -0,0 +1,5 @@
+numpy
+Pillow
+scikit-learn
+joblib
+datasets

svm_fashion.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51f932d8909a56d06dcf4dc6c8dc9454d2e8506949291352ca17e491d0db2b86
+size 133147243

svm_mnist.joblib

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:daee5169488a0e5eb3b481edfc1eaf86df62632f25e8335ae4e0ca268e826e93
+size 79199979