skilfoy/Check-my-progress-Deep-RL-Course
0
1import gradio as gr2from huggingface_hub import HfApi, hf_hub_download3from huggingface_hub.repocard import metadata_load4 5import pandas as pd6 7import requests8 9from utils import *10 11api = HfApi()12 13def get_user_models(hf_username, env_tag, lib_tag):14 """15 List the Reinforcement Learning models16 from user given environment and lib17 :param hf_username: User HF username18 :param env_tag: Environment tag19 :param lib_tag: Library tag20 """21 api = HfApi()22 models = api.list_models(author=hf_username, filter=["reinforcement-learning", env_tag, lib_tag])23 24 user_model_ids = [x.modelId for x in models]25 return user_model_ids26 27 28def get_user_sf_models(hf_username, env_tag, lib_tag):29 api = HfApi()30 models_sf = []31 models = api.list_models(author=hf_username, filter=["reinforcement-learning", lib_tag])32 33 user_model_ids = [x.modelId for x in models]34 35 for model in user_model_ids:36 meta = get_metadata(model)37 if meta is None:38 continue39 result = meta["model-index"][0]["results"][0]["dataset"]["name"]40 if result == env_tag:41 models_sf.append(model)42 43 return models_sf44 45 46def get_metadata(model_id):47 """48 Get model metadata (contains evaluation data)49 :param model_id50 """51 try:52 readme_path = hf_hub_download(model_id, filename="README.md")53 return metadata_load(readme_path)54 except requests.exceptions.HTTPError:55 # 404 README.md not found56 return None57 58 59def parse_metrics_accuracy(meta):60 """61 Get model results and parse it62 :param meta: model metadata63 """64 if "model-index" not in meta:65 return None66 result = meta["model-index"][0]["results"]67 metrics = result[0]["metrics"]68 accuracy = metrics[0]["value"]69 70 return accuracy71 72 73def parse_rewards(accuracy):74 """75 Parse mean_reward and std_reward76 :param accuracy: model results77 """78 default_std = -100079 default_reward= -100080 if accuracy != None:81 accuracy = str(accuracy)82 parsed = accuracy.split(' +/- ')83 if len(parsed)>1:84 mean_reward = float(parsed[0])85 std_reward = float(parsed[1])86 elif len(parsed)==1: #only mean reward 87 mean_reward = float(parsed[0])88 std_reward = float(0)89 else: 90 mean_reward = float(default_std)91 std_reward = float(default_reward)92 else:93 mean_reward = float(default_std)94 std_reward = float(default_reward)95 96 return mean_reward, std_reward97 98def calculate_best_result(user_model_ids):99 """100 Calculate the best results of a unit101 best_result = mean_reward - std_reward102 :param user_model_ids: RL models of a user103 """104 best_result = -1000105 best_model_id = ""106 for model in user_model_ids:107 meta = get_metadata(model)108 if meta is None:109 continue110 accuracy = parse_metrics_accuracy(meta)111 mean_reward, std_reward = parse_rewards(accuracy)112 result = mean_reward - std_reward113 if result > best_result:114 best_result = result115 best_model_id = model116 117 return best_result, best_model_id118 119def check_if_passed(model):120 """121 Check if result >= baseline122 to know if you pass123 :param model: user model124 """125 if model["best_result"] >= model["min_result"]:126 model["passed_"] = True127 128def certification(hf_username):129 results_certification = [130 {131 "unit": "Unit 1",132 "env": "LunarLander-v2",133 "library": "stable-baselines3",134 "min_result": 200,135 "best_result": 0,136 "best_model_id": "",137 "passed_": False138 },139 {140 "unit": "Unit 2",141 "env": "Taxi-v3",142 "library": "q-learning",143 "min_result": 4,144 "best_result": 0,145 "best_model_id": "",146 "passed_": False147 },148 {149 "unit": "Unit 3",150 "env": "SpaceInvadersNoFrameskip-v4",151 "library": "stable-baselines3",152 "min_result": 200,153 "best_result": 0,154 "best_model_id": "",155 "passed_": False156 },157 {158 "unit": "Unit 4",159 "env": "CartPole-v1",160 "library": "reinforce",161 "min_result": 350,162 "best_result": 0,163 "best_model_id": "",164 "passed_": False165 },166 {167 "unit": "Unit 4",168 "env": "Pixelcopter-PLE-v0",169 "library": "reinforce",170 "min_result": 5,171 "best_result": 0,172 "best_model_id": "",173 "passed_": False174 },175 {176 "unit": "Unit 5",177 "env": "ML-Agents-SnowballTarget",178 "library": "ml-agents",179 "min_result": -100,180 "best_result": 0,181 "best_model_id": "",182 "passed_": False183 },184 {185 "unit": "Unit 5",186 "env": "ML-Agents-Pyramids",187 "library": "ml-agents",188 "min_result": -100,189 "best_result": 0,190 "best_model_id": "",191 "passed_": False192 },193 {194 "unit": "Unit 6",195 "env": "PandaReachDense",196 "library": "stable-baselines3",197 "min_result": -3.5,198 "best_result": 0,199 "best_model_id": "",200 "passed_": False201 },202 {203 "unit": "Unit 7",204 "env": "ML-Agents-SoccerTwos",205 "library": "ml-agents",206 "min_result": -100,207 "best_result": 0,208 "best_model_id": "",209 "passed_": False210 },211 {212 "unit": "Unit 8 PI",213 "env": "LunarLander-v2",214 "library": "deep-rl-course",215 "min_result": -500,216 "best_result": 0,217 "best_model_id": "",218 "passed_": False219 },220 {221 "unit": "Unit 8 PII",222 "env": "doom_health_gathering_supreme",223 "library": "sample-factory",224 "min_result": 5,225 "best_result": 0,226 "best_model_id": "",227 "passed_": False228 },229 ] 230 231 for unit in results_certification:232 if unit["unit"] == "Unit 6":233 # Since Unit 6 can use PandaReachDense-v2 or v3234 user_models = get_user_models(hf_username, "PandaReachDense-v3", unit["library"])235 if len(user_models) == 0:236 print("Empty")237 user_models = get_user_models(hf_username, "PandaReachDense-v2", unit["library"])238 elif unit["unit"] != "Unit 8 PII":239 # Get user model240 user_models = get_user_models(hf_username, unit['env'], unit['library'])241 # For sample factory vizdoom we don't have env tag for now242 else: 243 user_models = get_user_sf_models(hf_username, unit['env'], unit['library'])244 245 # Calculate the best result and get the best_model_id246 best_result, best_model_id = calculate_best_result(user_models)247 248 # Save best_result and best_model_id249 unit["best_result"] = best_result250 unit["best_model_id"] = make_clickable_model(best_model_id)251 252 # Based on best_result do we pass the unit?253 check_if_passed(unit)254 unit["passed"] = pass_emoji(unit["passed_"])255 256 print(results_certification)257 258 df = pd.DataFrame(results_certification)259 df = df[['passed', 'unit', 'env', 'min_result', 'best_result', 'best_model_id']]260 return df261 262 263with gr.Blocks() as demo:264 gr.Markdown(f"""265 # 🏆 Check your progress in the Deep Reinforcement Learning Course 🏆266 You can check your progress here.267 268 - To get a certificate of completion, you must **pass 80% of the assignments**.269 - To get an honors certificate, you must **pass 100% of the assignments**.270 271 There's **no deadlines, the course is self-paced**.272 273 To pass an assignment your model result (mean_reward - std_reward) must be >= min_result274 275 **When min_result = -100 it means that you just need to push a model to pass this hands-on. No need to reach a certain result.**276 277 Just type your Hugging Face Username 🤗 (in my case skilfoy)278 """)279 280 hf_username = gr.Textbox(placeholder="skilfoy", label="Your Hugging Face Username")281 #email = gr.Textbox(placeholder="skilfoy@huggingface.co", label="Your Email (to receive your certificate)")282 check_progress_button = gr.Button(value="Check my progress")283 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"])284 check_progress_button.click(fn=certification, inputs=hf_username, outputs=output)285 286demo.launch()