<![CDATA[
================================================================================
* HTML page view: https://youngminpark2559.github.io/kaggle/bacteria-classification-at-the-genus-level/prj_root/README.html
================================================================================
* Introduction
- Competetion page: https://www.kaggle.com/c/bacteria-classification-at-the-genus-level
- Problem: multi class classification
- Data: Bacteria images + corresponding 4 class labels
- Goal: Train a classifier model using train dataset. Test the trained classifier by using test data
================================================================================
* Explanation on data
- 606 number of bacteria train images
- Sample train bacteria images
]]>
<![CDATA[
]]>
<![CDATA[
]]>
<![CDATA[
- Bacteria classes: ecoli, salmonella, staphylococus, listeria
================================================================================
* Libraries
- Python 3.6
- PyTorch 1.0.1.post2
- CUDA V10.0.130
- CuDNN v7.4
- Scikit-Learn
- And others which you can install whenever you run into unmet-dependencies
================================================================================
* Train information
- Tested epochs: 30
- Batch: 30
- Input image size: resized into (224,224)
- Network: Resnet with CBAM attention modules
- Loss function: Cross Entropy loss function
================================================================================
* Result
- Loss value decreasing
]]>
<![CDATA[
- Accuracy
(1) y_true
[array([3, 1, 2, 0, 3]) array([2, 2, 2, 1, 3]) array([2, 1, 0, 1, 0])
array([2, 2, 1, 3, 0]) array([0, 3, 1, 1, 1]) array([0, 0, 0, 2, 1])
array([2, 1, 2, 2, 3]) array([0, 1, 2, 0, 3]) array([2, 0, 2, 0, 3])
array([1, 2, 2, 3, 0]) array([1, 2, 0, 0, 0]) array([3, 0, 3, 0, 0])
array([2])]
(2) y_pred
[array([3, 1, 2, 0, 3]) array([2, 2, 2, 1, 3]) array([2, 1, 0, 1, 0])
array([2, 2, 1, 3, 0]) array([0, 3, 1, 1, 1]) array([0, 0, 0, 2, 1])
array([2, 0, 2, 2, 3]) array([0, 1, 2, 0, 3]) array([2, 0, 2, 0, 3])
array([1, 2, 2, 3, 0]) array([1, 2, 0, 0, 0]) array([3, 0, 3, 0, 0])
array([0])]
(3) Accuracy
0.9672131147540983
================================================================================
* To do
[--] 1. Test the trained classifier by using full test data
[--] 2. Make a submission and see the actual inference result
[--] 3. Implement other metrics (F1 score, precision, recall)
It was not simple to use metrics which are provided from scikit-leaarn,
because I guess I'm dealing with multi class problem instead of binary classification
So, I'm manually implementing metrics by using their fomulars.
]]>