import streamlit as st import pandas as pd import plotly.graph_objects as go import plotly.express as px import numpy as np from abc import ABC, abstractmethod # Set the title of the app st.title("Homer Simpson Meta-Learning with Hierarchical Reinforcement Learning Intrinsic Reward Lecture") # Display the image with a caption st.image("homer.webp", caption="Homer Simpson Meta-Learning HRL Lecture", use_column_width=True) # Display and play the audio files st.write("Audio Playback Meta-Learning with HRL Intrinsic Reward Lecture:") st.audio("h0.wav", format="audio/wav") st.audio("h1.wav", format="audio/wav") st.write("Oh, sweet Homer's doughnuts! If that second .wav file ain't playin', just download the darn thing! Mmm... downloads...") st.audio("h2.wav", format="audio/wav") st.image("intrinsic_reward_formulation.png", caption='Intrinsic Reward Formulation') st.write("Solving the first 5 equations @ (0,0):") df_0_0 = pd.read_csv('df_0_0.csv') st.write(df_0_0.shape) st.write("Example 1 via Method 1:") # Define parameters eta = 0.1 N_st = 1 epsilon = 1e-5 # Intrinsic reward formulation r_t_int = eta * (1 / (N_st + epsilon)**0.5) # Display the formulation with parameters plugged in st.latex(r""" r_{t}^{int} \eta \frac{1}{\sqrt{N(s_{t}) + \epsilon}} = 0.1 \frac{1}{\sqrt{1 + 1 x 10^{-5}}} """) st.write(f"Calculated intrinsic reward: {r_t_int}") st.write(f"Calculated intrinsic reward rounded 2 decimal places:", np.round(r_t_int,2)) st.dataframe(df_0_0[:1]) # Display the formulation with parameters plugged in st.write("Example 2 via Method 2:") st.latex(r""" r_{t}^{int} \eta \frac{1}{\sqrt{N(s_{t}) + \epsilon}} = 0.1 \frac{1}{\sqrt{2 + 1 x 10^{-5}}} """) # Abstract Base Class for Intrinsic Reward Calculation class IntrinsicRewardCalculator(ABC): @abstractmethod def calculate_intrinsic_reward(self, eta, count, epsilon): pass # Concrete Class for Intrinsic Reward Calculation class ConcreteIntrinsicRewardCalculator(IntrinsicRewardCalculator): def calculate_intrinsic_reward(self, eta, count, epsilon): return eta * (1 / np.sqrt(count + epsilon)) def populate_df_0_0(self, df_0_0, eta, count, epsilon): intrinsic_reward = self.calculate_intrinsic_reward(eta, count, epsilon) df_0_0.at[0, 'Intrinsic Reward'] = intrinsic_reward return df_0_0 # Example 2 parameters eta = 0.1 count = 2 epsilon = 1e-5 x,y = 0,0 # Create instance for Intrinsic Reward Calculation irc = ConcreteIntrinsicRewardCalculator() intrinsic_reward = irc.calculate_intrinsic_reward(0.1, 2, 1e-5) st.write(f"Intrinsic Reward @ {count} @ Coordinates {x,y}:", intrinsic_reward) st.write(f"Intrinsic Reward @ {count} @ Coordinates {x,y} rounded 4 decimal places:", np.round(intrinsic_reward,4)) # Populate the DataFrame with the calculated intrinsic reward df_0_0 = irc.populate_df_0_0(df_0_0, eta, count, epsilon) # Display the updated DataFrame st.dataframe(df_0_0[1:2]) st.write("Example 3 via Method 1:") # Example 3 parameters eta = 0.1 N_st = 3 epsilon = 1e-5 # Intrinsic reward formulation r_t_int = eta * (1 / (N_st + epsilon)**0.5) # Display the formulation with parameters plugged in st.latex(r""" r_{t}^{int} \eta \frac{1}{\sqrt{N(s_{t}) + \epsilon}} = 0.1 \frac{1}{\sqrt{3 + 1 x 10^{-5}}} """) st.write(f"Calculated intrinsic reward: {r_t_int}") st.write(f"Calculated intrinsic reward rounded 4 decimal places:", np.round(r_t_int,4)) st.dataframe(df_0_0[2:3]) # Display the formulation with parameters plugged in st.write("Example 4 via Method 2:") st.latex(r""" r_{t}^{int} \eta \frac{1}{\sqrt{N(s_{t}) + \epsilon}} = 0.1 \frac{1}{\sqrt{4 + 1 x 10^{-5}}} """) # Abstract Base Class for Intrinsic Reward Calculation class IntrinsicRewardCalculator(ABC): @abstractmethod def calculate_intrinsic_reward(self, eta, count, epsilon): pass # Concrete Class for Intrinsic Reward Calculation class ConcreteIntrinsicRewardCalculator(IntrinsicRewardCalculator): def calculate_intrinsic_reward(self, eta, count, epsilon): return eta * (1 / np.sqrt(count + epsilon)) def populate_df_0_0(self, df_0_0, eta, count, epsilon): intrinsic_reward = self.calculate_intrinsic_reward(eta, count, epsilon) df_0_0.at[0, 'Intrinsic Reward'] = intrinsic_reward return df_0_0 # Example 4 parameters eta = 0.1 count = 4 epsilon = 1e-5 x,y = 0,0 # Create instance for Intrinsic Reward Calculation irc = ConcreteIntrinsicRewardCalculator() intrinsic_reward = irc.calculate_intrinsic_reward(0.1, 4, 1e-5) st.write(f"Intrinsic Reward @ {count} @ Coordinates {x,y}:", intrinsic_reward) st.write(f"Intrinsic Reward @ {count} @ Coordinates {x,y} rounded 2 decimal places:", np.round(intrinsic_reward,2)) # Populate the DataFrame with the calculated intrinsic reward df_0_0 = irc.populate_df_0_0(df_0_0, eta, count, epsilon) # Display the updated DataFrame st.dataframe(df_0_0[3:4]) st.write("Example 5 via Method 1:") # Example 5 parameters eta = 0.1 N_st = 5 epsilon = 1e-5 # Intrinsic reward formulation r_t_int = eta * (1 / (N_st + epsilon)**0.5) # Display the formulation with parameters plugged in st.latex(r""" r_{t}^{int} \eta \frac{1}{\sqrt{N(s_{t}) + \epsilon}} = 0.1 \frac{1}{\sqrt{5 + 1 x 10^{-5}}} """) st.write(f"Calculated intrinsic reward: {r_t_int}") st.write(f"Calculated intrinsic reward rounded 4 decimal places:", np.round(r_t_int,4)) st.dataframe(df_0_0[4:5]) st.write("Oh, sweet Krusty-licious! At coordinates (0,0) for that plotly visualization, we need a whopping 7035 intrinsic reward calculations to get things rollin'! And don't forget to update those State Visitations. Those were just the first five. Mmm... 7030 more to go... D'oh!") # Define the grid and visitations grid = np.zeros((6, 6)) visitations = { (0, 0): 7035, (1, 0): 3579, (2, 0): 1359, (2, 1): 1707, (3, 1): 520, (4, 1): 227, (4, 2): 243, (5, 1): 217, (5, 2): 181, (5, 0): 241, (4, 0): 267, (5, 3): 179, (4, 3): 1034, (3, 3): 2163, (2, 3): 2080, (0, 1): 3313, (1, 1): 3015, (0, 2): 1846, (0, 3): 1104, (0, 4): 351, (1, 4): 518, (1, 3): 1497, (1, 2): 2236, (2, 2): 2239, (2, 4): 842, (1, 5): 238, (2, 5): 217, (0, 5): 341, (3, 5): 382, (4, 5): 1872, (4, 4): 2038, (3, 4): 1684, (3, 0): 383, (3, 2): 1102, (5, 4): 198 } # Fill the grid with visitations for (x, y), count in visitations.items(): grid[x, y] = count # Calculate the total number of visitations total_visitations = sum(visitations.values()) # Calculate the percentages percentages = {state: (count / total_visitations) * 100 for state, count in visitations.items()} # Print the percentages in the specified order order = [ (0, 0), (0, 1), (0, 2), (0, 3), (0, 4), (0, 5), (1, 0), (1, 1), (1, 2), (1, 3), (1, 4), (1, 5), (2, 0), (2, 1), (2, 2), (2, 3), (2, 4), (2, 5), (3, 0), (3, 1), (3, 2), (3, 3), (3, 4), (3, 5), (4, 0), (4, 1), (4, 2), (4, 3), (4, 4), (4, 5), (5, 0), (5, 1), (5, 2), (5, 3), (5, 4), (5, 5) ] st.title("State Visitations Visualization") st.write("### State Visitations Percentages:") for state in order: st.write(f"State {state}: {percentages.get(state, 0):.2f}%") # Create a pie chart labels = [f"State {state}" for state in visitations.keys()] values = list(visitations.values()) fig_pie = go.Figure(data=[go.Pie(labels=labels, values=values)]) fig_pie.update_layout(title_text="State Visitations Pie Chart") st.plotly_chart(fig_pie) # Create a heatmap fig_heatmap = px.imshow(grid, labels=dict(x="Column", y="Row", color="Visitations"), x=list(range(6)), y=list(range(6)), title="State Visitations Heatmap") fig_heatmap.update_xaxes(side="top") st.plotly_chart(fig_heatmap) # Load the CSV data df = pd.read_csv('goal_rows.csv') # Aggregate the data to count the number of visits to each State_2D visitation_counts = df['State_2D'].value_counts().reset_index() visitation_counts.columns = ['State_2D', 'Visitation_Count'] # Create the Plotly bar chart fig = px.bar(visitation_counts, x='State_2D', y='Visitation_Count', title='Goal Position Visitation Counts', labels={'State_2D': 'State 2D', 'Visitation_Count': 'Visitation Count'}) # Display the plot using Streamlit st.title('Goal Position Visitation Counts Visualization') st.plotly_chart(fig)