Download my_pages/rashomon_developer.py from verifiability-tutorial/verifiability: direct link, hf CLI and curl.
- Browser
- Download file 3.75 kB
-
https://huggingface.co/spaces/verifiability-tutorial/verifiability/resolve/main/my_pages/rashomon_developer.py
- Command line
-
hf download hf://spaces/verifiability-tutorial/verifiability/my_pages/rashomon_developer.py
-
curl -L -o rashomon_developer.py https://huggingface.co/spaces/verifiability-tutorial/verifiability/resolve/main/my_pages/rashomon_developer.py
3.75 kB
| import streamlit as st | |
| import matplotlib.pyplot as plt | |
| import numpy as np | |
| from utils import add_navigation, add_instruction_text, add_red_text | |
| from my_pages.rashomon_effect import plot_scatter | |
| from my_pages.rashomon_effect import income, credit, labels, colors | |
| plt.style.use('dark_background') | |
| def render(): | |
| add_navigation("txt_rashomon_developer", "txt_developer_decisions") | |
| add_instruction_text( | |
| """ | |
| Consider the same data as before. <br> | |
| Instead of directly choosing a model, you make development choices now. | |
| """ | |
| ) | |
| #### Choosing regularization | |
| st.markdown(""" | |
| **Regularization:** Regularization is a technique commonly used to stop AI models from learning the noise or small quirks in the data that might not generalize. | |
| Choose a regularization method: | |
| - L1 Regularization: Force your AI model to use less number of features, thus avoiding irrelevant features. | |
| - L2 Regularization: Force your AI model to rely less on each feature, even though you use all features, thus avoiding noisy dominance of any single feature. | |
| """ | |
| ) | |
| regularization_method = None | |
| if "regularization_method" in st.session_state: | |
| regularization_method = st.session_state.regularization_method | |
| col1, col2 = st.columns([1, 1]) | |
| with col1: | |
| if regularization_method=="l1": | |
| button_click_l1 = st.button("L1 Regularization", type="primary") | |
| else: | |
| button_click_l1 = st.button("L1 Regularization") | |
| if button_click_l1: | |
| st.session_state.regularization_method = "l1" | |
| st.rerun() | |
| with col2: | |
| if regularization_method=="l2": | |
| button_click_l2 = st.button("L2 Regularization", type="primary") | |
| else: | |
| button_click_l2 = st.button("L2 Regularization") | |
| if button_click_l2: | |
| st.session_state.regularization_method = "l2" | |
| st.rerun() | |
| #### Choosing random seed | |
| if regularization_method=="l1": | |
| st.markdown(""" | |
| **Randomness:** Sometimes there is randomness in the learning process. Let's flip a coin | |
| (You can just choose Heads or Tails, and we will assume we flipped a coin. It'll be our little secret). | |
| """ | |
| ) | |
| random_seed = None | |
| if "random_seed" in st.session_state: | |
| random_seed = st.session_state.random_seed | |
| col1, col2 = st.columns([1, 1]) | |
| with col1: | |
| if random_seed=="Heads": | |
| button_click_1 = st.button("Heads", type="primary") | |
| else: | |
| button_click_1 = st.button("Heads") | |
| if button_click_1: | |
| st.session_state.random_seed = "Heads" | |
| st.rerun() | |
| with col2: | |
| if random_seed=="Tails": | |
| button_click_2 = st.button("Tails", type="primary") | |
| else: | |
| button_click_2 = st.button("Tails") | |
| if button_click_2: | |
| st.session_state.random_seed = "Tails" | |
| st.rerun() | |
| #### Plot the final figure | |
| plot_chosen = None | |
| if regularization_method=="l2": | |
| plot_chosen = "slant" | |
| if regularization_method=="l1": | |
| if random_seed=="Heads": | |
| plot_chosen = "vertical" | |
| elif random_seed=="Tails": | |
| plot_chosen = "horizontal" | |
| if plot_chosen is not None: | |
| col1, col2, col3 = st.columns([1.5, 1, 1.5]) | |
| with col2: | |
| st.pyplot(plot_scatter(income, credit, colors, boundary_type=plot_chosen, highlight_point=None)) | |
| multiplicity_message = """ | |
| Your choices during model development lead you to this model. | |
| """ | |
| add_red_text(multiplicity_message) |