shahkv95 / evaluating-performance

We have determined that the flowernet network is 92% accurate (22/24) on the test data. You can see visually that the two misclassifications were both predicted to be bluebells, whereas they were actually a crocus and an iris. You could use standard MATLAB data analysis techniques to further investigate these misclassified images. A typical approach is to find which files contain the misclassified images, then import and view these images (or a subset of them) to see if there are any characteristics that are causing problems for the network. Note that this example used a much higher ratio of training images to test images than you would normally use in practice. This was done to reduce the amount of time needed to classify the test images in these interactions. If possible, in practice you should reserve enough test images so that the test results can be taken as representative of the network's behavior in general use.

---

---

---

Below is the roadmap of the Kush City with address and other information:

• 💻 Previously worked with Barclays UK - Data Platform and Business Intelligence team
• 🔭 Research project in Computational Finance with an intriguing interest in Fintech industry
• 🖌️ I used to create some designs in my spare time. Your design reviews at Pinterest would be valuable
• 📫 You can always find me at Linkedin
• 😄 Oh you wish to collaborate at Kaggle, ping me at Kaggle
• 🆘 If emails are what you are more comfortable with, then I think you would like to drop a mail at Kush email

---