8 PM – AI and Machine Learning – Joel (Gamas Sub)

Today We Did

1. Went over each students custom project and help them with the following in their custom kaggle project

   1. !pip install --force-reinstall \
        "numpy==1.26.4" \
        "scipy==1.15.3" \
        "matplotlib==3.7.2" \
        "fastai==2.7.19" \
        "torch==2.6.0" \
       "pillow==10.4.0"

   2. Most of them chose dataset for their projects where the object names is in the folder name instead of the file name, but their labeling functions assumes file_name is passed as the function parameter. And they are also using ImageDataLoaders.from_name_func instead of ImageDataLoaders.from_path_func. I assisted with all students to change their functions and to use ImageDataLoaders.from_path_func except for Aiden and Joshua because their datasets have both the name of the object in both folder and filename.
2. Setup Titanic Survival streamlit pycharm project and asked them to install requirements.txt

   1. fastai==2.8.4
      numpy==1.26.4
      scipy==1.13.1
      streamlit==1.52.1

Homework

Continue with your custom projects

FOR ALL STUDENTS

1. in your streamlit pycharm projects, make sure you add this file requirements.txt AND make sure it has the following content

   1. streamlit==1.40.1
      numpy==1.26.4
      scipy==1.14.1
      matplotlib==3.9.2
      pillow==10.4.0
      torch==2.6.0
      fastai==2.7.19

2. Open terminal and run

   1. pip install -r requirements.txt

For Individual Students

1. Aiden
   1. Train your model until you get small enough error_rate. Sometimes because of bad data set, small error_rate might not be possible.
   2. export your pkl file from Kaggle to your computer and start your Pycharm streamlit project.
2. Russell
   1. Fix your Kaggle labelling function so it can extract object name from the folder name instead of file name. Look at how we did this in single digit prediction project.
   2. Utilize ImageDataLoaders.from_path_func instead of from_name_func
   3. Export pkl file and start your Pycharm streamlit project.
3. Joshua
   1. Your project is pretty far (both Kaggle and Pycharm), which is good. Since we just fixed the fastai function in your Kaggle to 2.7.19, reexport your PKL file.
   2. Download the PKL file from Kaggle to your computer and to your Pycharm.
   3. Do some streamlit testing.
4. Rafael
   1. Fix your Kaggle labelling function so it can extract object name from the folder name instead of file name. Look at how we did this in single digit prediction project.
   2. Utilize ImageDataLoaders.from_path_func instead of from_name_func
   3. Export pkl file and start your Pycharm streamlit project.