8 thoughts on “CASE Kaufland, TEAM “Data Abusement Squad””
0
votes
The model itself and the algorithm behind it are very sound. However don’t copy and paste the case description as your business understanding and data analysis. The business understanding should be an answer from YOU to the following questions:
Who is this model for (e.g who are its end users)?
What do they want from the system?
In the data analysis section you need to show your understanding of the data not only its structure. Where there any problems with the data? Was there any noise? Did you have to do some preprocessing? Did you identify any patterns? What methods did you use to analyze the data? What algorithms or methods did you use to prepare your training, validation and testing sets?
Also add a model description section. In this section describe what models/technologies you used and why you decided to use them. This part is meant to help non-technical people such as the industry experts understand what you’ve actually made. People who are not familiar with Python won’t be able to understand what the system does just by looking at your code or its results.
The results section. Don’t just provide raw output provide some analysis of your system’s performance. List all methods and techniques you’ve used to evaluate your model. If you’ve made any graphs(confusion matrix, gain/lift charts) add them to make your article more presentable. You’ve given the accuracy score but what about other metrics (f-score, precision, recall)? Try to explain what the results tell you about the system. Does it confuse any objects (e.g does it classify apples as oranges sometimes)?
Don’t forget the job of the data scientist is not just to produce a good working model but to present it as well. Other than those remarks good job. 🙂
From the article for me it is not clear if the training data set provided by Kaufland is used to report the accuracy.
The models results are good and I particularly like the simplicity of the model!
Very well done!
However, there was no attempt to compare the results with similar model created with fully convolution network, that can be used directly with data set with different image sizes.
Hi, I think you had a bug in your code.
Your first epoch gives 98.5 accuracy and stays flat. Your loss is “binary crossentropy” and it should not be.
I am actually very new in this field (this is my first more complicated project), and I don’t know how the things actually work, yet… But the participation in this datathon makes me want to learn more 🙂
8 thoughts on “CASE Kaufland, TEAM “Data Abusement Squad””
The model itself and the algorithm behind it are very sound. However don’t copy and paste the case description as your business understanding and data analysis. The business understanding should be an answer from YOU to the following questions:
Who is this model for (e.g who are its end users)?
What do they want from the system?
In the data analysis section you need to show your understanding of the data not only its structure. Where there any problems with the data? Was there any noise? Did you have to do some preprocessing? Did you identify any patterns? What methods did you use to analyze the data? What algorithms or methods did you use to prepare your training, validation and testing sets?
Also add a model description section. In this section describe what models/technologies you used and why you decided to use them. This part is meant to help non-technical people such as the industry experts understand what you’ve actually made. People who are not familiar with Python won’t be able to understand what the system does just by looking at your code or its results.
The results section. Don’t just provide raw output provide some analysis of your system’s performance. List all methods and techniques you’ve used to evaluate your model. If you’ve made any graphs(confusion matrix, gain/lift charts) add them to make your article more presentable. You’ve given the accuracy score but what about other metrics (f-score, precision, recall)? Try to explain what the results tell you about the system. Does it confuse any objects (e.g does it classify apples as oranges sometimes)?
Don’t forget the job of the data scientist is not just to produce a good working model but to present it as well. Other than those remarks good job. 🙂
From the article for me it is not clear if the training data set provided by Kaufland is used to report the accuracy.
The models results are good and I particularly like the simplicity of the model!
Very well done!
However, there was no attempt to compare the results with similar model created with fully convolution network, that can be used directly with data set with different image sizes.
Hi, I think you had a bug in your code.
Your first epoch gives 98.5 accuracy and stays flat. Your loss is “binary crossentropy” and it should not be.
Good job! The only thing I am missing is trying to use a pre-trained model like VGG.
I was thinking about trying to add VGG, but I wanted to invest more of my time in trying to understand why and how things work.
Hi, I think you had a bug in your code.
Your first epoch gives 98.5 accuracy and stays flat. Your loss is “binary crossentropy” and it should not be.
I had those moments myself , nothing to worry about. you are on a good track just use categorical end redo it you may still get nice results. Cheers
I am actually very new in this field (this is my first more complicated project), and I don’t know how the things actually work, yet… But the participation in this datathon makes me want to learn more 🙂