import gradio as gr
from huggingface_hub import HfApi, hf_hub_download
from huggingface_hub.repocard import metadata_load
import pandas as pd
import requests
from utils import *
api = HfApi()
def get_user_models(hf_username, env_tag, lib_tag):
"""
List the Reinforcement Learning models
from user given environment and lib
:param hf_username: User HF username
:param env_tag: Environment tag
:param lib_tag: Library tag
"""
api = HfApi()
models = api.list_models(author=hf_username, filter=["reinforcement-learning", env_tag, lib_tag])
user_model_ids = [x.modelId for x in models]
return user_model_ids
def get_user_sf_models(hf_username, env_tag, lib_tag):
api = HfApi()
models_sf = []
models = api.list_models(author=hf_username, filter=["reinforcement-learning", lib_tag])
user_model_ids = [x.modelId for x in models]
for model in user_model_ids:
meta = get_metadata(model)
if meta is None:
continue
result = meta["model-index"][0]["results"][0]["dataset"]["name"]
if result == env_tag:
models_sf.append(model)
return models_sf
def get_metadata(model_id):
"""
Get model metadata (contains evaluation data)
:param model_id
"""
try:
readme_path = hf_hub_download(model_id, filename="README.md")
return metadata_load(readme_path)
except requests.exceptions.HTTPError:
# 404 README.md not found
return None
def parse_metrics_accuracy(meta):
"""
Get model results and parse it
:param meta: model metadata
"""
if "model-index" not in meta:
return None
result = meta["model-index"][0]["results"]
metrics = result[0]["metrics"]
accuracy = metrics[0]["value"]
return accuracy
def parse_rewards(accuracy):
"""
Parse mean_reward and std_reward
:param accuracy: model results
"""
default_std = -1000
default_reward= -1000
if accuracy != None:
accuracy = str(accuracy)
parsed = accuracy.split(' +/- ')
if len(parsed)>1:
mean_reward = float(parsed[0])
std_reward = float(parsed[1])
elif len(parsed)==1: #only mean reward
mean_reward = float(parsed[0])
std_reward = float(0)
else:
mean_reward = float(default_std)
std_reward = float(default_reward)
else:
mean_reward = float(default_std)
std_reward = float(default_reward)
return mean_reward, std_reward
def calculate_best_result(user_model_ids):
"""
Calculate the best results of a unit
best_result = mean_reward - std_reward
:param user_model_ids: RL models of a user
"""
best_result = -1000
best_model_id = ""
for model in user_model_ids:
meta = get_metadata(model)
if meta is None:
continue
accuracy = parse_metrics_accuracy(meta)
mean_reward, std_reward = parse_rewards(accuracy)
result = mean_reward - std_reward
if result > best_result:
best_result = result
best_model_id = model
return best_result, best_model_id
def check_if_passed(model):
"""
Check if result >= baseline
to know if you pass
:param model: user model
"""
if model["best_result"] >= model["min_result"]:
model["passed_"] = True
def certification(hf_username):
results_certification = [
{
"unit": "Unit 1",
"env": "LunarLander-v2",
"library": "stable-baselines3",
"min_result": 200,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 2",
"env": "Taxi-v3",
"library": "q-learning",
"min_result": 4,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 3",
"env": "SpaceInvadersNoFrameskip-v4",
"library": "stable-baselines3",
"min_result": 200,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 4",
"env": "CartPole-v1",
"library": "reinforce",
"min_result": 350,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 4",
"env": "Pixelcopter-PLE-v0",
"library": "reinforce",
"min_result": 5,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 5",
"env": "ML-Agents-SnowballTarget",
"library": "ml-agents",
"min_result": -100,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 5",
"env": "ML-Agents-Pyramids",
"library": "ml-agents",
"min_result": -100,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 6",
"env": "PandaReachDense",
"library": "stable-baselines3",
"min_result": -3.5,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 7",
"env": "ML-Agents-SoccerTwos",
"library": "ml-agents",
"min_result": -100,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 8 PI",
"env": "LunarLander-v2",
"library": "deep-rl-course",
"min_result": -500,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
{
"unit": "Unit 8 PII",
"env": "doom_health_gathering_supreme",
"library": "sample-factory",
"min_result": 5,
"best_result": 0,
"best_model_id": "",
"passed_": False
},
]
for unit in results_certification:
if unit["unit"] == "Unit 6":
# Since Unit 6 can use PandaReachDense-v2 or v3
user_models = get_user_models(hf_username, "PandaReachDense-v3", unit["library"])
if len(user_models) == 0:
print("Empty")
user_models = get_user_models(hf_username, "PandaReachDense-v2", unit["library"])
elif unit["unit"] != "Unit 8 PII":
# Get user model
user_models = get_user_models(hf_username, unit['env'], unit['library'])
# For sample factory vizdoom we don't have env tag for now
else:
user_models = get_user_sf_models(hf_username, unit['env'], unit['library'])
# Calculate the best result and get the best_model_id
best_result, best_model_id = calculate_best_result(user_models)
# Save best_result and best_model_id
unit["best_result"] = best_result
unit["best_model_id"] = make_clickable_model(best_model_id)
# Based on best_result do we pass the unit?
check_if_passed(unit)
unit["passed"] = pass_emoji(unit["passed_"])
print(results_certification)
df = pd.DataFrame(results_certification)
df = df[['passed', 'unit', 'env', 'min_result', 'best_result', 'best_model_id']]
return df
with gr.Blocks() as demo:
gr.Markdown("""
# 🏆 Progress in the Deep Reinforcement Learning Course 🏆
(From Thomassimonini)
""")
# Hidden Textbox to pass the username
hf_username = gr.Textbox(value="yesbut", visible=False)
output = gr.components.Dataframe(value= certification(hf_username), headers=["Pass?", "Unit", "Environment", "Baseline", "Your best result", "Your best model id"], datatype=["markdown", "markdown", "markdown", "number", "number", "markdown", "bool"])
# Automatically load progress
demo.load(fn=certification, inputs=[hf_username], outputs=output)
demo.launch()