app.py

import gradio as gr
import joblib
import pandas as pd
import datasets
import json
import numpy as np

# Load the model
pipe = joblib.load("./model.pkl")

title = "Premium Amount Prediction"
description = "This model predicts the Premium Amount. Drag and drop any slice from the dataset or edit values as you wish in the dataframe component below."

# Load and prepare dataset
df = datasets.load_dataset("silvaKenpachi/mental_health")["train"].to_pandas()
df.dropna(axis=0, inplace=True)

# Load configuration
with open("./config.json") as f:
    config_dict = json.load(f)
all_headers = config_dict["sklearn"]["columns"]

# Filter headers to only include those present in the dataset
headers = [col for col in all_headers if col in df.columns]

# Define input and output interfaces
#inputs = [gr.Dataframe(headers=headers, row_count=(2, "dynamic"), col_count=(len(headers), "fixed"), label="Input Data", interactive=True)]

#working code that returns only 2 rows in output
#inputs = [gr.Dataframe(headers=all_headers, row_count=(2, "dynamic"), col_count=(len(all_headers), "fixed"), label="Input Data", interactive=True)]
#outputs = [gr.Dataframe(row_count=(2, "dynamic"), col_count=(1, "fixed"), label="Predictions", headers=["Depression"])]

# Define input and output interfaces with dynamic row counts
inputs = [gr.Dataframe(
    headers=headers,
    row_count=(10, "dynamic"),  # Use tuple format (min_rows, "dynamic")
    col_count=(len(headers), "fixed"),
    label="Input Data",
    interactive=True
)]

outputs = [gr.Dataframe(
    row_count=(10, "dynamic"),  # Use tuple format (min_rows, "dynamic")
    col_count=(2, "fixed"),
    label="Predictions",
    headers=["Name", "Depression"]
)]


#def infer(inputs):
    #data = pd.DataFrame(inputs, columns=headers)
    #predictions = pipe.predict(data)
    #return pd.DataFrame(predictions, columns=["Depression"])

#code to fix missing columns with na
#def infer(inputs):
    #data = pd.DataFrame(inputs, columns=headers)
    # Add missing columns with default values (e.g., 0)
    #for col in all_headers:
        #if col not in data.columns:
            #data[col] = 0
    # Ensure the order of columns matches the training data
    #data = data[all_headers]
    #predictions = pipe.predict(data)
    #return pd.DataFrame(predictions, columns=["Depression"])


#def infer(inputs):
    #data = pd.DataFrame(inputs, columns=headers)
    
    # Replace empty strings with NaN
    #data = data.replace('', np.nan)
    
    # Add missing columns with default values (e.g., 0)
    #for col in all_headers:
        #if col not in data.columns:
            #data[col] = 0
    
    # Ensure the order of columns matches the training data
    #data = data[all_headers]
    
    # Fill NaN values with default values (e.g., 0)
    #data = data.fillna(0)
    
    # Convert all data to float
    #data = data.astype(float)
    
    #predictions = pipe.predict(data)
    #return pd.DataFrame(predictions, columns=["Name", "Depression"])
    #return pd.DataFrame({
        #'Name': data['Name'],
        #'Depression': predictions
    #})

def infer(inputs):
    # Create DataFrame from inputs
    data = pd.DataFrame(inputs, columns=headers)
    
    # Create a copy of the input DataFrame to preserve original data
    prediction_data = data.copy()
    
    # Replace empty strings with NaN for numeric columns only
    numeric_columns = [col for col in all_headers if col != 'Name']
    prediction_data[numeric_columns] = prediction_data[numeric_columns].replace('', np.nan)
    
    # Add missing columns with default values
    for col in all_headers:
        if col not in prediction_data.columns:
            prediction_data[col] = 0
    
    # Ensure the order of columns matches the training data
    prediction_data = prediction_data[all_headers]
    
    # Fill NaN values in numeric columns only
    prediction_data[numeric_columns] = prediction_data[numeric_columns].fillna(0)
    
    # Convert numeric columns to float
    prediction_data[numeric_columns] = prediction_data[numeric_columns].astype(float)
    
    # Make predictions
    predictions = pipe.predict(prediction_data)
    
    # Create output DataFrame using original names
    return pd.DataFrame({
        'Name': data['Name'],
        'Depression': predictions
    })




gr.Interface(
    fn=infer,
    inputs=inputs,
    outputs=outputs,
    title=title,
    description=description,
    examples=[df[headers].head(3).values.tolist()],
    cache_examples=False
).launch(debug=True)