app.py

import gradio as gr
import pandas as pd
import os
import pickle
from datetime import datetime
from sklearn import set_config
set_config(transform_output="pandas")

# PAGE CONFIG
# page_icon = "💐"

# Setup variables and constants
# datetime.now().strftime('%d-%m-%Y _ %Hh %Mm %Ss')
DIRPATH = os.path.dirname(os.path.realpath(__file__))
tmp_dir = os.path.join(DIRPATH, "src", "assets", "tmp",)
os.system(f'rm -vf {tmp_dir}/*.csv')
tmp_df_fp = os.path.join(
    tmp_dir, f"history_{datetime.now().strftime('%d-%m-%Y _ %Hh %Mm %Ss')}.csv")
ml_core_fp = os.path.join(DIRPATH, "src", "assets",
                          "ml", "crop_recommandation2.pkl")
init_df = pd.DataFrame(
    {
        "N": [],
        "P": [],
        "K": [],
        "temperature": [],
        "humidity": [],
        "ph": [],
        "rainfall": [],
    }
)

# FUNCTIONS


def load_ml_components(fp):
    "Load the ml component to re-use in app"
    with open(fp, "rb") as f:
        object = pickle.load(f)
    return object


def setup(fp):
    "Setup the required elements like files, models, global variables, etc"

    # history frame
    if not os.path.exists(fp):
        df_history = init_df.copy()
    else:
        df_history = pd.read_csv(fp)

    df_history.to_csv(fp, index=False)

    return df_history


def select_categorical_widget(col_index, col_name, encoder):
    """This function will return the right widget to use for each categorical feature
    """

    categories = encoder.categories_[col_index].tolist()
    n_unique = len(categories)

    # print(
    #     f"[Info] unique categories for feature {col_name}  ({type(categories)}) are : {categories}")

    if n_unique == 2:
        print(
            f"[Info] unique categories for feature '{col_index}' {col_name}  ({type(categories)}) are : {categories}")

        widget = gr.Checkbox(label=f"Enter {col_name}", value=categories)
    elif n_unique <= 5:
        widget = gr.Radio(label=f"Enter {col_name}", choices=categories)
    else:
        widget = gr.Dropdown(label=f"Enter {col_name}", choices=categories)

    return widget


def make_prediction(*args):
    """Function that takes values from fields to make 1-by-1 prediction
    """
    print(
        f"[Info] input args of the function {args} ")
    raw = {k: [val if not isinstance(val, list) else val[0]]
           for val, k in zip(args, num_cols+cat_cols)}
    print(
        f"[Info] input modified a bit {raw}\n")

    df_input = pd.DataFrame(raw)
    global df_history

    # print(f"\n[Info] Input information as dataframe: \n{df_input.to_string()}")
    df_input.drop_duplicates(inplace=True, ignore_index=True)
    print(f"\n[Info] Input with duplicated rows: \n{df_input.to_string()}")

    df_input_num, df_input_cat = None, None

    if len(cat_cols) > 0:
        df_input_cat = df_input[cat_cols].copy()
        if cat_imputer:
            df_input_cat = cat_imputer.transform(df_input_cat)
        if encoder:
            df_input_cat = encoder.transform(df_input_cat)

    if len(num_cols) > 0:
        df_input_num = df_input[num_cols].copy()
        if num_imputer:
            df_input_num = num_imputer.transform(df_input_num)
        if scaler:
            df_input_num = scaler.transform(df_input_num)

    df_input_ok = pd.concat([df_input_num, df_input_cat], axis=1)

    prediction_output = model.predict_proba(df_input_ok)

    output = model.predict_proba(df_input_ok)

    # store confidence score/ probability for the predicted class
    confidence_score = output.max(axis=-1)
    df_input["confidence score"] = confidence_score

    # get index of the predicted class
    predicted_idx = output.argmax(axis=-1)

    # store index then replace by the matching label
    df_input["predicted crop"] = predicted_idx
    predicted_label = df_input["predicted crop"].replace(idx_to_labels)
    df_input["predicted crop"] = predicted_label

    print(
        f"[Info] Prediction output (of type '{type(prediction_output)}') from passed input: {prediction_output} of shape {prediction_output.shape}")

    # print(f"[Info] Prediction: {prediction_output}")
    # df_input['prediction'] = prediction_output

    print(
        f"\n[Info] output information as dataframe: \n{df_input.to_string()}")
    df_history = pd.concat([df_history, df_input], ignore_index=True).drop_duplicates(
        ignore_index=True, keep='last')
    df_history.to_csv(tmp_df_fp, index=False, )

    return df_input


def download():
    return gr.File.update(label="History File",
                          visible=True,
                          value=tmp_df_fp)


def hide_download():
    return gr.File.update(label="History File",
                          visible=False)


# Setup execution
ml_components_dict = load_ml_components(fp=ml_core_fp)

num_cols = [
    "N",
    "P",
    "K",
    "temperature",
    "humidity",
    "ph",
    "rainfall",
]
cat_cols = ml_components_dict['cat_cols'] if 'cat_cols' in ml_components_dict else [
]
num_imputer = ml_components_dict['num_imputer'].set_output(transform="pandas") if (
    'num_cols' in ml_components_dict and 'num_imputer' in ml_components_dict) else None
cat_imputer = ml_components_dict['cat_imputer'].set_output(transform="pandas") if (
    'cat_cols' in ml_components_dict and 'cat_imputer' in ml_components_dict) else None
scaler = ml_components_dict['scaler'].set_output(
    transform="pandas") if 'scaler' in ml_components_dict else None
encoder = ml_components_dict['encoder'] if 'encoder' in ml_components_dict else None
model = ml_components_dict['model']
labels = ml_components_dict['labels'] if 'labels' in ml_components_dict else []
idx_to_labels = {i: l for (i, l) in enumerate(labels)}
end2end_pipeline = ml_components_dict['pipeline']
print(f"\n[Info] ML components loaded: {list(ml_components_dict.keys())}")

df_history = setup(tmp_df_fp)


# APP Interface

# Main page
demo_inputs = []
with gr.Blocks() as demo:
    gr.Markdown('''<img class="center" src="https://www.verdict.co.uk/wp-content/uploads/2018/12/Agri-tech.jpg"  width="60%" height="60%">
    <style>
    .center {
    display: block;
    margin-left: auto;
    margin-right: auto;
    width: 50%;
    }
    </style>''')
    gr.Markdown('''<center><h1> 🚀 Agri-Tech App 🚀 </h1><center>''')
    gr.Markdown('''
    This is a ML API for classification of crop to plant on a land regarding some features
    ''')

    # multiple_of = 7
    # print(f"\n[INFO] {init_df.shape[1]}\n")

    with gr.Row():
        for i in range(0, init_df.shape[1],):
            demo_inputs.append(gr.Number(label=f"Enter {num_cols[i]}"))

    # with gr.Row():
    #     for i in range(0,init_df.shape[1], multiple_of):
    #         gr.Number(label=f"Enter {num_cols[i]}")
    #         gr.Number(label=f"Enter {num_cols[i+1]}")
    #         # gr.Number(label=f"Enter {num_cols[i+2]}")
    output = gr.Dataframe(df_history)

    btn_predict = gr.Button("Predict")
    btn_predict.click(fn=make_prediction, inputs=demo_inputs, outputs=output)

    file_obj = gr.File(label="History File",
                       visible=False
                       )

    btn_download = gr.Button("Download")
    btn_download.click(fn=download, inputs=[], outputs=file_obj)
    output.change(fn=hide_download, inputs=[], outputs=file_obj)

# second demo
# num_inputs = [gr.Number(label=f"Enter {col}") for col in num_cols]
# cat_inputs = [select_categorical_widget(col_index=i, col_name=col, encoder=encoder) for i, col in enumerate(cat_cols)]

# inputs = num_inputs + cat_inputs

# demo = gr.Interface(
#     make_prediction,
#     inputs,
#     "dataframe", # "number"
#     examples=[
#         # [2, "cat", ["Japan", "Pakistan"], "park", ["ate", "swam"], True],
#     ]
# )


if __name__ == "__main__":
    demo.launch(share=True, debug=True)