18/01/2025
Teaching My AI the Nepali Way🤖: Chaos, Momo, and Everything in Between
Training ML Model..?? It’s like teaching a kid to separate momos 🍲 from samosas and chowmein 🍜. Here's how I trained my “Nepali kid” AI with CIFAR-10 Dataset. Buckle up, it’s chaotic, like Kathmandu traffic.!! 🚦
1. Data Prep: Tarkari Pasal Style 🛒
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
Normalising data = washing tarkari before cooking. Clean data = better model! 🥦
2. Building Dimaag 🧠: Nepali Thali
def create_model():
model = models.Sequential([
layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)),
layers.MaxPooling2D((2, 2)),
layers.Flatten(),
layers.Dense(64, activation='relu'),
layers.Dense(10, activation='softmax')
])
return model
Conv2D = rice 🍚, MaxPooling = achar 🥒, Dense = tarkari. Nepali AI thali!
3. Data Augmentation: Gym for AI 🏋️
data_augmentation = tf.keras.Sequential([
layers.RandomFlip("horizontal"),
layers.RandomRotation(0.1),
layers.RandomZoom(0.1)
])
Augmentation = teaching AI chaos. Like finding roadblocks on the way to Thamel.
In a Hauba Way : AI learns to recognise a dai in daura suruwal or jeans. Nepali versatility at its best.!! 🎉
4. Training: AI’s School Life 📚
for epoch in range(epochs):
for step, (x_batch_train, y_batch_train) in enumerate(datagen.flow(x_train, y_train)):
with tf.GradientTape() as tape:
logits = model(x_batch_train, training=True)
loss = loss_fn(y_batch_train, logits)
Training = school. Mistakes.?? Loads. Progress..?? Slow. Sometimes, it blanks out like a kid in class.😅
5. Nepali Twist: CutMix📸
def cutmix(image, label):
lam = np.random.beta(1.0, 1.0)
rand_index = np.random.randint(0, x_train.shape[0])
image[bbx1:bbx2, bby1:bby2, :] = x_train[rand_index][bbx1:bbx2, bby1:bby2, :]
return image, lam * label + (1 - lam) * y_train[rand_index]
CutMix is chaos, like mixing momos and chowmein on one plate and asking AI to figure it out. 🤔
6. Validation: Nosy Relatives Mode🎓
val_logits = model(x_batch_val, training=False)
val_acc_metric.update_state(y_batch_val, val_logits)
Validation = report card time. If it flops here, it's like scoring 30 after claiming you studied all night. 👀
7. Testing: SEE Results 🏆
test_loss, test_accuracy = model.evaluate(x_test, y_test)
print(f"Test accuracy: {test_accuracy:.4f}")
Testing AI is like an exam. It nails cats 🐱 and planes ✈️, but calls a frog 🐸 a dog 🐕. Classic AI fail.! 😂 Anyone else’s AI make these funny mistakes.?? Share your fails! 🚀
Lessons Learnt
1. Clean Data: Like washing veggies for tarkari, don’t skip it! 🥦
2. Embrace Chaos: Augmentation = handling KTM traffic jams. 🚦
3. Prove It: Validation is like impressing nosy relatives, don’t fail! 😂👀
AI is chaos and creativity, like coding during load-shedding with chiya tough but rewarding.
