TensorFlow architecture (eager vs graph execution) - Model Approaches Compared

Model Pipeline - TensorFlow architecture (eager vs graph execution)

This pipeline shows how TensorFlow processes data using two main modes: eager execution, which runs operations immediately like a calculator, and graph execution, which builds a plan first and then runs it for speed.

Data Flow - 7 Stages

1Input Data

1000 rows x 10 features→Raw data loaded into TensorFlow→1000 rows x 10 features

[[0.5, 1.2, ..., 0.3], [0.7, 0.8, ..., 0.1], ...]

↓

2Preprocessing

1000 rows x 10 features→Normalize features (subtract mean, divide by std)→1000 rows x 10 features

[[-0.1, 0.5, ..., -0.3], [0.2, -0.1, ..., -0.5], ...]

↓

3Feature Engineering

1000 rows x 10 features→Add polynomial features (square each feature)→1000 rows x 20 features

[[-0.1, 0.5, ..., -0.3, 0.01, 0.25, ..., 0.09], [0.2, -0.1, ..., -0.5, 0.04, 0.01, ..., 0.25], ...]

↓

4Model Training (Eager Mode)

800 rows x 20 features→Run operations immediately for each batch→Model weights updated after each batch

Weights tensor updated step-by-step

↓

5Model Training (Graph Mode)

800 rows x 20 features→Build computation graph, then run all at once→Model weights updated after full graph run

Graph executed, weights updated

↓

6Evaluation

200 rows x 20 features→Calculate loss and accuracy on test set→Loss scalar, accuracy scalar

Loss=0.25, Accuracy=0.88

↓

7Prediction

1 row x 20 features→Run model to predict output→1 row x 1 prediction

[0.76]

Training Trace - Epoch by Epoch

Loss
1.0 |*       
0.8 | *      
0.6 |  *     
0.4 |   *    
0.2 |    *   
0.0 +---------
      1 2 3 4 5
      Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.85	0.60	Initial training with eager execution, loss high, accuracy low
2	0.60	0.75	Loss decreased, accuracy improved
3	0.45	0.82	Model learning well, eager mode stable
4	0.35	0.87	Loss continues to drop, accuracy rises
5	0.30	0.90	Good convergence, eager execution effective

Prediction Trace - 3 Layers

Layer 1: Input Layer

Layer 2: Dense Layer with ReLU (Eager Mode)

Layer 3: Dense Layer with Sigmoid (Graph Mode)

Model Quiz - 3 Questions

What is the main difference between eager and graph execution in TensorFlow?

A Eager requires less memory than graph

B Eager runs operations immediately; graph builds a plan first

C Graph execution cannot be used for training

D Eager mode is slower because it uses graphs

Key Insight

TensorFlow's eager execution mode is intuitive and easy to debug because it runs operations immediately, like doing math step-by-step. Graph execution builds a full plan before running, which can be faster for big tasks. Watching loss go down and accuracy go up during training shows the model is learning well.