dnisąpoliczone

app.py

@@ -1,172 +1,176 @@
-import streamlit as st
-import wandb
-import pandas as pd
-import os
-import time
-from utils import fetch_runs_to_df, fetch_run, fetch_models_to_df
-
-### WANDB
-
-# Access the API key from the environment variable
-wandb_api_key = os.getenv('WANDB_API_KEY')
-
-# Log in to wandb using the API key
-if wandb_api_key:
- wandb.login(key=wandb_api_key)
-else:
- st.error("WANDB_API_KEY not found in environment variables.")
-
-# Initialize W&B API
-api = wandb.Api()
-
-### STREAMLIT APP
-
-# Define available projects (bookmarks)
-projects = {
- "Competition 1": {"entity": "urbaniak-bruno-safescanai", "project": "pytorch-intro", "description": "This is a beginner-friendly competition using PyTorch."},
- "Competition 2": {"entity": "urbaniak-bruno-safescanai", "project": "basic-intro", "description": "This competition focuses on basic introduction to AI models."},
- "Competition 3 (mamymodelexd)": {"entity": "urbaniak-bruno-safescanai", "project": "simple-cnn", "description": "This is an advanced competition for CNN model training."},
- "Competition 4": {"entity": "urbaniak-bruno-safescanai", "project": "model-validation", "description": "This competition covers techniques for model validation."},
- "Competition 5": {"entity": "safe-scan-ai", "project": "melanoma-1", "description": "This competition focuses on melanoma detection."},
- # Add more projects as needed
-}
-
-# Initialize leader info
-@st.cache_data
-def update_leader_info(leader_info, competition, best_model):
- if leader_info.get(competition) is None:
- leader_info[competition] = {
- "Username": best_model["Username"],
- "Model Name": best_model["Model Name"],
- "Hotkey": best_model["Hotkey"],
- "Date": time.strftime("%Y-%m-%d"),
- "UID": "xd",
- "Days on Top": 1
- }
- else:
- if leader_info[competition]["UID"] == "xd": #best_model["ID"]:
- leader_info[competition]["Days on Top"] += 1
- else:
- leader_info[competition]["Username"] = best_model["Username"]
- leader_info[competition]["Model Name"] = best_model["Model Name"]
- leader_info[competition]["Hotkey"] = best_model["Hotkey"]
- leader_info[competition]["Date"] = time.strftime("%Y-%m-%d")
- leader_info[competition]["UID"] = best_model["ID"]
- leader_info[competition]["Days on Top"] = 1
- return leader_info[competition]
-
-
-@st.cache_data()
-def load_competition_data(last_update_time=None):
- ranking_dfs = {}
- for competition, details in projects.items():
- df = fetch_runs_to_df(api, projects, competition)
- if not df.empty:
- # Convert to DataFrame
- ranking_df = fetch_models_to_df(api, projects, competition, df)
-
- # Rank the models by accuracy, then recall, then ROC AUC
- rank_by = ["Recall", "Accuracy", "ROC AUC"]
- ascending = [False, False, False]
- ranking_df = ranking_df.sort_values(by=rank_by, ascending=ascending)
- 
- ranking_dfs[competition] = ranking_df
- 
- # Update the timestamp of the last update
- last_update_time = time.time()
- return ranking_dfs, last_update_time
-
-
-# Streamlit app main function
-def main():
- # Set Streamlit page configuration to wide
- st.set_page_config(layout="wide")
- st.title("LaUltimate Dashboard")
- st.subheader("Welcome to the Competition Dashboard!")
- st.write("Explore the various AI competitions and their respective rankings. Select a competition to view more details and rankings.")
-
- # Define the update interval (e.g., 10 minutes)
- update_interval = 4 * 60 # 10 minutes in seconds
-
- # Initialize session state to track the last update time
- if 'last_update_time' not in st.session_state:
- st.session_state.last_update_time = None
- # Initialize or update leader_info
- if "leader_info" not in st.session_state:
- st.session_state.leader_info = {}
-
- # Check if we need to refresh the data
- if st.session_state.last_update_time is None or (time.time() - st.session_state.last_update_time > update_interval):
- ranking_dfs, st.session_state.last_update_time = load_competition_data(st.session_state.last_update_time)
-
- # Update leader info for each competition
- for competition in ranking_dfs:
- best_model = ranking_dfs[competition].iloc[0]
- st.session_state.leader_info[competition] = update_leader_info(st.session_state.leader_info, competition, best_model)
- 
- else:
- ranking_dfs, _ = load_competition_data(st.session_state.last_update_time)
-
- # Display the time of the last update
- st.write(f"Last updated: {time.ctime(st.session_state.last_update_time)}")
-
- st.markdown("### Competitions")
-
- st.write("### Select Competition")
-
- # Create a header for the table with additional columns
- cols = st.columns([1, 3, 2, 2, 2, 2, 1, 2]) # Adjust column width ratios if needed
- cols[0].write("Index")
- cols[1].write("Competition Name")
- cols[2].write("Leader")
- cols[3].write("Model")
- cols[4].write("Date")
- cols[5].write("UID")
- cols[6].write("Hotkey")
- cols[7].write("Days on Top")
-
-
- # Display the project names in a table-like structure with buttons and placeholders for additional data
- for index, (competition, details) in enumerate(projects.items(), start=1):
- best_model = ranking_dfs[competition].iloc[0]
-
- # Use the leader_info from session_state
- leader_info = st.session_state.get("leader_info")
-
-
- cols = st.columns([1, 3, 2, 2, 2, 2, 1, 2])
- cols[0].write(index)
- if cols[1].button(competition):
- st.experimental_set_query_params(comp=competition)
- cols[2].write(leader_info[competition]["Username"])
- cols[3].write(leader_info[competition]["Model Name"])
- cols[4].write(leader_info[competition]["Date"])
- cols[5].write(leader_info[competition]["UID"])
- cols[6].write(leader_info[competition]["Hotkey"])
- cols[7].write(leader_info[competition]["Days on Top"])
-
-
- # Get query parameters to identify which competition page to show
- query_params = st.experimental_get_query_params() 
-
- if "comp" in query_params:
- competition_name = query_params["comp"][0]
- st.write(f"Selected competition: {competition_name}")
- competition_details = projects.get(competition_name, {})
- description = competition_details.get("description", "No description available.")
- 
- st.header(f"Competition: {competition_name}")
- st.write(description)
-
- df = ranking_dfs[competition_name]
- if not df.empty:
- st.dataframe(df)
- else:
- st.warning("No runs available for ranking.")
- else:
- st.write("Please select a competition to view details.")
-
-
-# Run the app
-if __name__ == "__main__":
- main()
+import streamlit as st
+import wandb
+import pandas as pd
+import os
+import time
+from utils import fetch_runs_to_df, fetch_run, fetch_models_to_df
+
+### WANDB
+
+# Access the API key from the environment variable
+wandb_api_key = os.getenv('WANDB_API_KEY')
+
+# Log in to wandb using the API key
+if wandb_api_key:
+ wandb.login(key=wandb_api_key)
+else:
+ st.error("WANDB_API_KEY not found in environment variables.")
+
+# Initialize W&B API
+api = wandb.Api()
+
+### STREAMLIT APP
+
+# Define available projects (bookmarks)
+projects = {
+ "Competition 1": {"entity": "urbaniak-bruno-safescanai", "project": "pytorch-intro", "description": "This is a beginner-friendly competition using PyTorch."},
+ "Competition 2": {"entity": "urbaniak-bruno-safescanai", "project": "basic-intro", "description": "This competition focuses on basic introduction to AI models."},
+ "Competition 3 (mamymodelexd)": {"entity": "urbaniak-bruno-safescanai", "project": "simple-cnn", "description": "This is an advanced competition for CNN model training."},
+ "Competition 4": {"entity": "urbaniak-bruno-safescanai", "project": "model-validation", "description": "This competition covers techniques for model validation."},
+ "Competition 5": {"entity": "safe-scan-ai", "project": "melanoma-1", "description": "This competition focuses on melanoma detection."},
+ # Add more projects as needed
+}
+
+# Initialize leader info
+@st.cache_data
+def update_leader_info(leader_info, competition, best_model):
+ if leader_info.get(competition) is None:
+ leader_info[competition] = {
+ "Username": best_model["Username"],
+ "Model Name": best_model["Model Name"],
+ "Hotkey": best_model["Hotkey"],
+ "Date": time.strftime("%Y-%m-%d"),
+ "UID": "xd",
+ "Days on Top": 1
+ }
+ else:
+ if leader_info[competition]["UID"] == "xd": #best_model["ID"]:
+ # update the days on top to the difference between the current date and the date of the last update
+ current_date = time.strftime("%Y-%m-%d")
+ last_update_date = leader_info[competition]["Date"]
+ days_on_top = (current_date - last_update_date).days
+ leader_info[competition]["Days on Top"] = days_on_top
+ else:
+ leader_info[competition]["Username"] = best_model["Username"]
+ leader_info[competition]["Model Name"] = best_model["Model Name"]
+ leader_info[competition]["Hotkey"] = best_model["Hotkey"]
+ leader_info[competition]["Date"] = time.strftime("%Y-%m-%d")
+ leader_info[competition]["UID"] = best_model["ID"]
+ leader_info[competition]["Days on Top"] = 1
+ return leader_info[competition]
+
+
+@st.cache_data()
+def load_competition_data(last_update_time=None):
+ ranking_dfs = {}
+ for competition, details in projects.items():
+ df = fetch_runs_to_df(api, projects, competition)
+ if not df.empty:
+ # Convert to DataFrame
+ ranking_df = fetch_models_to_df(api, projects, competition, df)
+
+ # Rank the models by accuracy, then recall, then ROC AUC
+ rank_by = ["Recall", "Accuracy", "ROC AUC"]
+ ascending = [False, False, False]
+ ranking_df = ranking_df.sort_values(by=rank_by, ascending=ascending)
+ 
+ ranking_dfs[competition] = ranking_df
+ 
+ # Update the timestamp of the last update
+ last_update_time = time.time()
+ return ranking_dfs, last_update_time
+
+
+# Streamlit app main function
+def main():
+ # Set Streamlit page configuration to wide
+ st.set_page_config(layout="wide")
+ st.title("LaUltimate Dashboard")
+ st.subheader("Welcome to the Competition Dashboard!")
+ st.write("Explore the various AI competitions and their respective rankings. Select a competition to view more details and rankings.")
+
+ # Define the update interval (e.g., 10 minutes)
+ update_interval = 4 * 60 # 10 minutes in seconds
+
+ # Initialize session state to track the last update time
+ if 'last_update_time' not in st.session_state:
+ st.session_state.last_update_time = None
+ # Initialize or update leader_info
+ if "leader_info" not in st.session_state:
+ st.session_state.leader_info = {}
+
+ # Check if we need to refresh the data
+ if st.session_state.last_update_time is None or (time.time() - st.session_state.last_update_time > update_interval):
+ ranking_dfs, st.session_state.last_update_time = load_competition_data(st.session_state.last_update_time)
+
+ # Update leader info for each competition
+ for competition in ranking_dfs:
+ best_model = ranking_dfs[competition].iloc[0]
+ st.session_state.leader_info[competition] = update_leader_info(st.session_state.leader_info, competition, best_model)
+ 
+ else:
+ ranking_dfs, _ = load_competition_data(st.session_state.last_update_time)
+
+ # Display the time of the last update
+ st.write(f"Last updated: {time.ctime(st.session_state.last_update_time)}")
+
+ st.markdown("### Competitions")
+
+ st.write("### Select Competition")
+
+ # Create a header for the table with additional columns
+ cols = st.columns([1, 3, 2, 2, 2, 2, 1, 2]) # Adjust column width ratios if needed
+ cols[0].write("Index")
+ cols[1].write("Competition Name")
+ cols[2].write("Leader")
+ cols[3].write("Model")
+ cols[4].write("Date")
+ cols[5].write("UID")
+ cols[6].write("Hotkey")
+ cols[7].write("Days on Top")
+
+
+ # Display the project names in a table-like structure with buttons and placeholders for additional data
+ for index, (competition, details) in enumerate(projects.items(), start=1):
+ best_model = ranking_dfs[competition].iloc[0]
+
+ # Use the leader_info from session_state
+ leader_info = st.session_state.get("leader_info")
+
+
+ cols = st.columns([1, 3, 2, 2, 2, 2, 1, 2])
+ cols[0].write(index)
+ if cols[1].button(competition):
+ st.experimental_set_query_params(comp=competition)
+ cols[2].write(leader_info[competition]["Username"])
+ cols[3].write(leader_info[competition]["Model Name"])
+ cols[4].write(leader_info[competition]["Date"])
+ cols[5].write(leader_info[competition]["UID"])
+ cols[6].write(leader_info[competition]["Hotkey"])
+ cols[7].write(leader_info[competition]["Days on Top"])
+
+
+ # Get query parameters to identify which competition page to show
+ query_params = st.experimental_get_query_params() 
+
+ if "comp" in query_params:
+ competition_name = query_params["comp"][0]
+ st.write(f"Selected competition: {competition_name}")
+ competition_details = projects.get(competition_name, {})
+ description = competition_details.get("description", "No description available.")
+ 
+ st.header(f"Competition: {competition_name}")
+ st.write(description)
+
+ df = ranking_dfs[competition_name]
+ if not df.empty:
+ st.dataframe(df)
+ else:
+ st.warning("No runs available for ranking.")
+ else:
+ st.write("Please select a competition to view details.")
+
+
+# Run the app
+if __name__ == "__main__":
+ main()