Introduction to TensorFlow / PyTorch

You don't write Neural Networks from scratch in production. You use a framework. The two leaders are TensorFlow (by Google) and PyTorch (by Meta).

1. Tensors 📦

A Tensor is just a fancy name for a multi-dimensional array (like a NumPy array), but it can run on a GPU.

• Scalar: 0-D Tensor (a single number)
• Vector: 1-D Tensor
• Matrix: 2-D Tensor
• 3D Tensor: e.g., an RGB image (Height x Width x 3)

2. PyTorch: The Researcher's Choice 🔬

PyTorch is loved for its "pythonic" style and dynamic computation graph. It feels like writing standard Python.

import torch
import torch.nn as nn

# Define a simple network
model = nn.Sequential(
    nn.Linear(10, 50),  # Input layer (10 features) -> Hidden (50 neurons)
    nn.ReLU(),          # Activation
    nn.Linear(50, 1)    # Hidden -> Output (1 prediction)
)

# Forward pass
input_data = torch.randn(1, 10)
output = model(input_data)
print(output)

3. TensorFlow / Keras: The Production Choice 🏭

TensorFlow (via its high-level API Keras) is great for deployment and production pipelines.

import tensorflow as tf
from tensorflow.keras import layers

# Define the same network
model = tf.keras.Sequential([
    layers.Dense(50, activation='relu', input_shape=(10,)),
    layers.Dense(1)
])

# Compile
model.compile(optimizer='adam', loss='mse')

4. The Training Loop 🔄

In PyTorch, you write the training loop manually (great for control). In Keras, you just call .fit().

PyTorch Loop:

1. optimizer.zero_grad() (Clear gradients)
2. output = model(x) (Forward)
3. loss = loss_fn(output, y) (Calculate error)
4. loss.backward() (Backprop)
5. optimizer.step() (Update weights)

Interactive Challenge: PyTorch Basics

PyTorch is installed in this environment! Let's do some tensor math.

Quiz

Key Takeaways

✅ Tensors are GPU-ready arrays.
✅ PyTorch is flexible and great for research.
✅ Keras is high-level and easy to use.

What's Next?

Now that we can build a basic network, let's build one that can "see".

Next Chapter: Convolutional Neural Networks (CNNs).