app.py

# app.py
import os
import pandas as pd
import zipfile
from huggingface_hub import hf_hub_download, HfApi
from autogluon.tabular import TabularPredictor
import gradio as gr
import numpy as np
MODEL_REPO = "samder03/2025-24679-tabular-autolguon-predictor"
LOCAL_DIR = "/tmp/autogluon_predictor"
os.makedirs(LOCAL_DIR, exist_ok=True)

# Download the predictor zip from HF
zip_path = hf_hub_download(
    repo_id=MODEL_REPO,
    filename="autogluon_predictor_dir.zip",
    local_dir=LOCAL_DIR
)

# Extract the full predictor directory
with zipfile.ZipFile(zip_path, 'r') as zip_ref:
    zip_ref.extractall(LOCAL_DIR)

# Load the predictor from the extracted folder
predictor = TabularPredictor.load(LOCAL_DIR, require_py_version_match=False)
print("Predictor loaded successfully!")
print("Label column:", predictor.label)
print("Feature columns:", predictor.feature_metadata.get_features())

# Build Gradio input widgets
features = predictor.feature_metadata.get_features()
inputs = []
input_names = []

sets = ["Journey Together","Journey Together","Destined Rivals","Destined Rivals","Stellar Crown","Journey Together","Evolutions","Evolutions","Primal Clash","Primal Clash","Pokemon 151","Prismatic Evolutions",
"Pokemon 151","Pokemon 151","Prismatic Evolutions","Paldea Evolved","White Flare","White Flare","Paradox Rift","Fates Collide","Evolutions","Evolutions","Next Destinies",
"Base Set","Generations","Generations","XY","Arceus","Roaring Skies","BREAKpoint","EX Sandstorm","Ancient Origins","Double Crisis","McDonalds"]
sets_un = np.unique(sets)
print('unique sets:', sets_un)
print(type(sets_un))

desired_order = ["Card", "Year", "Card Set", "Artwork Style", "Condition", "Market Value", "Set Number Eq"]
for c in desired_order:
  input_names.append(c)
  if c=="Card":
    inputs.append(gr.Textbox(value="", label=c))
  elif c=="Year":
    inputs.append(gr.Number(value=2021, precision=0, label=c))
  elif c=="Card Set":
    inputs.append(gr.Dropdown(choices=sets_un.tolist(), value="Journey Together", label=c))
  elif c=="Artwork Style":
    inputs.append(gr.Dropdown(choices=["Full Art", "Standard", "Holo", "Reverse Holo", "Full Art Gold", "Full Art Gold", "Promo"], value="Full Art", label=c))
  elif c=="Condition":
    inputs.append(gr.Dropdown(choices=["Mint", "Near Mint", "Lightly Played", "Heavily Played"], value="Mint", label=c))
  elif c=="Market Value":
    inputs.append(gr.Number(label=c, value=0.0))
  elif c=="Set Number Eq":
    inputs.append(gr.Number(label=c, value = 0.0))

examples_dicts = [
    {
        "Card": "Pikachu V",
        "Year": 2021,
        "Card Set": "Journey Together",
        "Artwork Style": "Full Art",
        "Condition": "Mint",
        "Market Value": 50.0,
        "Set Number Eq": 0.6
    },
    {
        "Card": "Charizard",
        "Year": 1999,
        "Card Set": "Base Set",
        "Artwork Style": "Holo",
        "Condition": "Near Mint",
        "Market Value": 12.0,
        "Set Number Eq": 1.4
    }
]

# Convert examples dicts into positional lists matching input_names
examples = [[ex[name] for name in input_names] for ex in examples_dicts]

def predict_record(*args):
  record = {name: val for name, val in zip(input_names, args)}
  df_in = pd.DataFrame([record])
  return str(predictor.predict(df_in).iloc[0])

iface = gr.Interface(
  fn=predict_record,
  inputs=inputs,
  outputs=gr.Textbox(label="Is this card a collector's item?"),
  title="Pokémon Card Collector's Item Predictor (AutoGluon)",
  description="Predicts whether a Pokémon card is a collector's item.",
  examples=examples
  )


if __name__ == "__main__":
  iface.launch()