Deep Learning Tasks in Python (.ipynb/Google Colab)
Budget: £20 – £250 GBP
1. Fine-Grained Classification:
The fine-Grained Image Classification tasks distinguish subtle differences between classes. In both academia and industry, it is highly important to get the best possible performance out of any model. The objective here is to train a network that can accurately classify models of airplanes. The dataset comes with a training set, a validation set and a test set. You are only allowed to train your model using the training set. Having a validation set is helpful to optimise your model hyper-parameters. The test set is only used to generate the final accuracy values. You should not use the test set for training under any circumstances. This is a .csv file which I will provide.
You can use any model architecture you want. You may use any of the seminal classification architectures or you can design your own network. It is recommended that you use transfer learning (by using weights pre-trained on ImageNet). You can try to freeze some layers of your network during fine-tuning, or you can fine tune the entire model.
You should find the best hyper-parameters you can, so you get the best results possible (this is what the validation set is for).
At the end, after model training is complete, you should assess the performance of your classifier using the test set and report Accuracy, Precision and Recall. You should also create a Confusion Matrix.
2. Time-Series Model:
Much of the data we need to process is time-series data. In this part of the task we give you twelve months’ worth of data for the logins and logouts of students to a mythical university set of computers. The file is a comma separated value (CSV) file with the following columns:
• Date and time of the event
• The event type – either LOGIN or LOGOUT
• The cluster on which the event occurred
• The duration – for a LOGIN this is the number of milli-seconds the user was logged in for, for a LOGOUT this is zero
• The total number of users logged in at that point in time
The fine-Grained Image Classification tasks distinguish subtle differences between classes. In both academia and industry, it is highly important to get the best possible performance out of any model. The objective here is to train a network that can accurately classify models of airplanes. The dataset comes with a training set, a validation set and a test set. You are only allowed to train your model using the training set. Having a validation set is helpful to optimise your model hyper-parameters. The test set is only used to generate the final accuracy values. You should not use the test set for training under any circumstances. This is a .csv file which I will provide.
You can use any model architecture you want. You may use any of the seminal classification architectures or you can design your own network. It is recommended that you use transfer learning (by using weights pre-trained on ImageNet). You can try to freeze some layers of your network during fine-tuning, or you can fine tune the entire model.
You should find the best hyper-parameters you can, so you get the best results possible (this is what the validation set is for).
At the end, after model training is complete, you should assess the performance of your classifier using the test set and report Accuracy, Precision and Recall. You should also create a Confusion Matrix.
2. Time-Series Model:
Much of the data we need to process is time-series data. In this part of the task we give you twelve months’ worth of data for the logins and logouts of students to a mythical university set of computers. The file is a comma separated value (CSV) file with the following columns:
• Date and time of the event
• The event type – either LOGIN or LOGOUT
• The cluster on which the event occurred
• The duration – for a LOGIN this is the number of milli-seconds the user was logged in for, for a LOGOUT this is zero
• The total number of users logged in at that point in time
Related categories:
Python
Software Architecture
Machine Learning (ML)
Neural Networks
Deep Learning