Skip to content

In this notebook I will try to recreate a curious model used in an App shown in the HBO serie "Sillicon Valley".

License

Notifications You must be signed in to change notification settings

YeyoM/Not_Hotdog_Model

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Not_Hotdog_Model

In this notebook I will try to recreate a curious model used in an App shown in the HBO serie "Sillicon Valley".

For this model, I will use the hotdog-nothotdog dataset from Kaggle, and to build the model my way to go was to use Tensorflow using Keras layers to build a Convolutional Neural Network and a pre-trained model to make use of transfer learning from the mobileNet v2 model.

Along the notebook and the code blocks, I will try to explain as best as I can what the building process of this model works and at the end obviously the results I got.

Hopefully I can also build the App as in the serie with this model I built, which is the first model I built by my own.

A detailed explanation of how the model was built take a look at the notebook attached in this repo.

You can found the google colab notebook in the following link:

Here are the results I got.

Results

Training

For the training process I run it for 20 epochs, with a batch size of 32 and a learning rate of 0.0001. Using the mobileNet v2 and some more layers including dropout and image augmentation to prevent overfitting. The results I got were the following:

training

Evaluation

For the evaluation process I used the validation set of the dataset and the results I got were the following:

evaluate

Predictions

For the predictions I used the test set of the dataset and the results I got were the following:

predictions

About

In this notebook I will try to recreate a curious model used in an App shown in the HBO serie "Sillicon Valley".

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published