Creating a branch and push to check.(dummy)

1. Create an environment: -conda env list

2. Create an environment and activate: -if you see base at first means you have successfully installed conda -conda create -n python= -conda activate env_name

3. Deactivate environment -conda deactivate

4. Create a directory for work -mkdir digit-classification -code . --> open visual code -save your work to exp.py file

5. Validate the files

6. Run python file and validate.

7. So we should create a requirement.txt to pass the required installment params -package_name==version_number

   -pip install -r requirement.txt

8. See the version of any lib installed -pip list | grep matplotlib

9. Running the py file again -python exp.py (if all packages installed correctly , it should run)

10. Change params , model

11. GIT : a. avoid git add . b. git add file_names c. git commit d. git push

---

system requirements: OS h/w -- may be skipped -- general commodity h/w is required

how to setup: install conda

conda create -n digits python=3.9 conda activate digits pip install -r requirements.txt

how to run

python exp.py

Meaning of failure:

poor performance metrics

coding runtime/compile error

the model gave bad predictions on the new test samples during demo.

feature:

vary model hyper parameters

Overview of train test split : 100 samples 2 class classification/binary classification : image of carrot or turnip 50 samples : carrots | 50 samples : turnips | This kind of data distribution : balanced/uniform

x amount data for training
n-x amount of data for testing

70 samples for training : 35 carrots , 35 turnips
30 samples for testing  : 12 carrots , 15 turnips

hence,
calculate some eval metric (train model (70 samples for training : 35 carrots , 35 turnips),(30 samples for testing  : 12 carrots , 15 turnips))==performance

In practice : train , development/validation , test

train = training the model(model type , model hyperparameters , model iterations)
dev = selecting the model
test = reportig the performance