import pandas as pd
import json
from sklearn.model_selection import train_test_split

# Load JSON data
with open('data.json', 'r', encoding='utf-8') as f:
    data = json.load(f)

# Extract the primary data structure
names = list(data.keys())
images = [item.get('images', []) for item in data.values()]

# Collect details into separate columns dynamically using a dictionary comprehension
details_columns = {key: [] for key in {k for v in data.values() for k in v.get('details', {})}}

for value in data.values():
    for detail_key in details_columns:
        details_columns[detail_key].append(value.get('details', {}).get(detail_key))

# Create the DataFrame
columns = {
    "Name": names,
    "Images": images,
    **details_columns
}
df = pd.DataFrame(columns)

# Generate HTML column for multiple images
def generate_html(image_paths):
    # Convert string representation of list to an actual list
    if isinstance(image_paths, str):
        image_paths = eval(image_paths)  # Convert stringified list to Python list
    elif not isinstance(image_paths, list):
        return None  # Handle unexpected cases
    # Generate HTML for all image paths
    return ''.join([f'<img src="{path}" style="width:100px; margin:5px;" />' for path in image_paths])

# Add HTML column
df['Images_HTML'] = df['Images'].apply(generate_html)

# Flatten the dataset so each image gets its own row

# Remove columns with all None values
df_flattened = df.dropna(axis=1, how='all')

# Remove column named 'Name' if it exists
df_flattened = df_flattened.drop(columns=['Charging portsLog in to see.'], errors='ignore')

df_flattened = df_flattened.loc[:, df_flattened.notna().sum() >= 10]
# Split data into train, test, and validation sets

df_flattened.to_parquet('train_with_html.parquet', index=False)

print("Data has been flattened, split, and saved to 'train.csv', 'test.csv', and 'val.csv'.")
print("Train dataset with HTML column saved to 'train_with_html.parquet'.")