import gradio as gr
import pandas as pd
import pickle
import numpy as np
from tensorflow import keras

treemodel = pickle.load(open('decision_tree.pkl', 'rb'))
nnmodel = keras.models.load_model("nnmodel.h5")

def onehot(df, column):
    df = df.copy()
    dummies = pd.get_dummies(df[column], prefix='type')
    df = pd.concat([df,dummies], axis=1)
    df = df.drop(column, axis=1)
    return df


def dataframe(df):
    df = onehot(df, column='type')

    #if the 'type' column doesn't have value 'CASH_OUT' then add a column 'type_CASH_OUT' with value 0
    if 'type_CASH_OUT' not in df.columns:
        df['type_CASH_OUT'] = 0
    #if the 'type' column doesn't have value 'TRANSFER' then add a column 'type_TRANSFER' with value 0
    if 'type_TRANSFER' not in df.columns:
        df['type_TRANSFER'] = 0
    #if the 'type' column doesn't have value 'PAYMENT' then add a column 'type_PAYMENT' with value 0
    if 'type_PAYMENT' not in df.columns:
        df['type_PAYMENT'] = 0
    #if the 'type' column doesn't have value 'DEBIT' then add a column 'type_DEBIT' with value 0
    if 'type_DEBIT' not in df.columns:
        df['type_DEBIT'] = 0
    #if the 'type' column doesn't have value 'PAYMENT' then add a column 'type_PAYMENT' with value 0
    if 'type_PAYMENT' not in df.columns:
        df['type_PAYMENT'] = 0
    
    df = df.drop(['nameOrig','nameDest','isFraud'], axis=1)
    return df

def tree(file_obj):
    df = pd.read_csv(file_obj.name)
    df = dataframe(df)
    y_pred = treemodel.predict(df)
    pred_df = pd.DataFrame(y_pred, columns = ['predictedFraud'])
    #append the predictions to the original dataframe
    df_original = pd.read_csv(file_obj.name)
    pred_df = pd.concat([df_original, pred_df], axis=1)
    return pred_df

def nn(file_obj):
    nn_df = pd.read_csv(file_obj.name)
    nn_df = dataframe(nn_df)
    y_prednn = nnmodel.predict(nn_df)
    pred_proc = pd.DataFrame(y_prednn, columns = ['predictedFraudProbability'])
    pred=np.where(y_prednn<0.44,0,1)
    pred_dfnn = pd.DataFrame(pred, columns = ['predictedFraud'])
    #append the predictions to the original dataframe
    df_originalnn = pd.read_csv(file_obj.name)
    pred_dfnn = pd.concat([df_originalnn, pred_proc, pred_dfnn], axis=1)
    return pred_dfnn

file = gr.components.File(file_count="single", type="file", label="Fisierul CSV cu tranzactii", optional=False)
tree_output = gr.components.Dataframe(max_rows=20, max_cols=None, overflow_row_behaviour="paginate", type="pandas", label="predictedFraud - Predictii bazate pe modelul de clasificare DECISION TREE")
nn_output = gr.components.Dataframe(max_rows=20, max_cols=None, overflow_row_behaviour="paginate", type="pandas", label="predictedFraud - Predictii bazate pe modelul de clasificare NEURAL NETWORK")
tree_interface = gr.Interface(
    fn=tree, 
    inputs=file, 
    outputs=tree_output,
)

nn_interface = gr.Interface(
    fn=nn, 
    inputs=file, 
    outputs=nn_output,
)

def report_tree(file_obj):
    creport_tree = pd.read_csv('report_tree.csv')
    return creport_tree

def report_nn(file_obj):
    creport_nn = pd.read_csv('report_nn.csv')
    return creport_nn

cr_tree = gr.components.DataFrame(type="pandas", label="Classification Report (Decision Tree)")
cr_nn = gr.components.DataFrame(type="pandas", label="Classification Report (Neural Network)")

r_tree = gr.Interface(
    fn=report_tree,
    inputs = file,
    outputs = cr_tree,
    css="body {background-color: #f5f5f5; width: 100%; height: 100%; margin: 0; padding: 0;}"
)

r_nn = gr.Interface(
    fn=report_nn,
    inputs = file,
    outputs = cr_nn,
    css="body {background-color: #f5f5f5; width: 100%; height: 100%; margin: 0; padding: 0;}"
)

#tree_interface.launch(inline=True)
gr.Parallel(tree_interface, nn_interface, r_tree, r_nn).launch(inline=True)