# Evaluation
In order to assess whether a certain image is compliant, your model/approach independent of its architecture should answer the following questions/sub-tasks:

- Are there 2 screws visible in the image?
- Are there 2 tie-wraps visible in the image?
- Are there 2 rubbers visible in the image?
- Are there 2 plastic covers visible in the image?
- Are there 2 foams visible in the image?

As [Kaggle ](https://www.kaggle.com/) doesn't support custom metrics, we couldn't choose several metrics to indicate the performance of your model/approach in each category. Eventually, all the predictions are treated in the same way regardless of their sub-task. [micro-f1-score](https://en.wikipedia.org/wiki/F-score) which takes class frequency into account, is used to evaluate your model/approach. At this point, class means an outcome of `0` or `1`. 

# Submission File

The submission file is required to be in a CSV format. The file should contain a header with `image_id_and_sub_task` and `prediction` columns and have the following format:
- `image_id_and_sub_task` column contains the image id and the sub-task information related to the image in `<image_id>_<sub-task>` format.
- `prediction` column contains `0` or `1` in case of the absence or presence of the related object, respectively.
-  **Pay attention, 5 rows related to 5 sub-tasks are required for each image.**
 
```
image_id_and_sub_task,prediction
image_id_0_two_screws,1
image_id_0_two_tie_wraps,1
image_id_0_two_rubbers,1
image_id_0_two_plastic_covers,1
image_id_0_two_foams,1
image_id_1_two_screws,0
image_id_1_two_tie_wraps,0
image_id_1_two_rubbers,0
image_id_1_two_plastic_covers,1
image_id_1_two_foams,1
...
```