How can Tensorflow be used with the flower dataset to compile and fit the model?

In this article, we will learn how can we compile a model and fit the flower dataset to it. TO fit a dataset on a model we need to first create a data pipeline, create the model’s architecture using TensorFlow high-level API, and then before fitting the model on the data using data pipelines we need to compile the model with an appropriate loss function and optimizer and a metric to understand the whether the model is making progress epoch after epoch or not.

Importing Libraries

Python libraries make it very easy for us to handle the data and perform typical and complex tasks with a single line of code.

• Pandas – This library helps to load the data frame in a 2D array format and has multiple functions to perform analysis tasks in one go.
• Numpy – Numpy arrays are very fast and can perform large computations in a very short time.
• Matplotlib – This library is used to draw visualizations.
• Sklearn – This module contains multiple libraries having pre-implemented functions to perform tasks from data preprocessing to model development and evaluation.
• Tensorflow – This is an open-source library that is used for Machine Learning and Artificial intelligence and provides a range of functions to achieve complex functionalities with single lines of code.

Now, let’s check the total number of images we have across all the classes of flowers. The link to the dataset is here https://www.kaggle.com/datasets/alxmamaev/flowers-recognition.

Output:

4317

Output:

 

Output:

 

Let’s check the different classes present in the training data and which class has been assigned to which integer.

Output:

array(['daisy', 'dandelion', 'rose', 'sunflower', 'tulip'], dtype=object)

Data Visualization

In this section, we will try to understand and visualize some images which have been provided to us to build the classifier for each class. Also, we will check for the imbalance problem.

Output:

A pie chart to visualize the data distribution

From the above graph, we can say that there is a little data imbalance problem in the given dataset. But handling the data balance is not an objective of this article.

Output:

Some images from the training dataset

Output:

((3669,), (648,))

Now by using the above function we will be implementing our training data input pipeline and the validation data pipeline.

Model Development

We will use pre-trained weight for an Inception network which is trained on imagenet dataset. This dataset contains millions of images for around 1000 classes of images. The parameters of a model we import are already trained on millions of images and for weeks so, we do not need to train them again.

Output:

94765736/94765736 [==============================] - 5s 0us/step

Model Architecture

We will implement a model using the  Functional API of Keras which will contain the following parts:

• The base model is the Inception model in this case.
• The Flatten layer flattens the output of the base model’s output.
• Then we will have two fully connected layers followed by the output of the flattened layer.
• We have included some BatchNormalization layers to enable stable and fast training and a Dropout layer before the final layer to avoid any possibility of overfitting.
• The final layer is the output layer which outputs soft probabilities for the three classes. 

While compiling a model we provide these three essential parameters:

• optimizer – This is the method that helps to optimize the cost function by using gradient descent.
• loss – The loss function by which we monitor whether the model is improving with training or not.
• metrics – This helps to evaluate the model by predicting the training and the validation data.

Now we are ready to train our model.

Output:

Epoch 1/5
115/115 [==============================] - 8s 60ms/step - loss: 1.5825 - auc: 0.7000 - val_loss: 1.6672 - val_auc: 0.7152
Epoch 2/5
115/115 [==============================] - 7s 59ms/step - loss: 1.3806 - auc: 0.7650 - val_loss: 1.4497 - val_auc: 0.7531
Epoch 3/5
115/115 [==============================] - 8s 68ms/step - loss: 1.2619 - auc: 0.7980 - val_loss: 1.3494 - val_auc: 0.7751
Epoch 4/5
115/115 [==============================] - 7s 58ms/step - loss: 1.1828 - auc: 0.8242 - val_loss: 1.3371 - val_auc: 0.7751
Epoch 5/5
115/115 [==============================] - 7s 60ms/step - loss: 1.0954 - auc: 0.8485 - val_loss: 1.8526 - val_auc: 0.7215

In the below code, we will create a data frame from the log obtained from the training of the model.

Output:

 

Let’s visualize the training loss and the validation loss of the data.

Output:

Training loss v/s Validation loss

Let’s visualize the training AUC and the validation AUC of the data.

Output:

Training AUC v/s Validation AUC