app.py

import json
import os

import joblib
import matplotlib.pyplot as plt
import numpy as np
import streamlit as st
from matplotlib.colors import ListedColormap
from tensorflow.keras.models import load_model

from utils.utils import input_predict

directory = "utils"
model_1_path = os.path.join(directory, "nnmodel_117d.keras")
model_2_path = os.path.join(directory, "nnmodel_2d.keras")
pca_adim_path = os.path.join(directory, "pca_2d.pkl")
json_file_path = os.path.join(directory, "columns.json")
mushroom_attributes_path = os.path.join(directory, "mushroom_attributes.json")
prediction_grid_path = os.path.join(directory, "pred_grid.npz")

model_1 = load_model(model_1_path)
model_2 = load_model(model_2_path)
pca_2d_loaded = joblib.load(pca_adim_path)
with open(json_file_path, 'r') as json_file:
    df_columns = json.load(json_file)
with open(mushroom_attributes_path, 'r') as json_file:
    loaded_dicts = json.load(json_file)
with np.load(prediction_grid_path) as data:
    xx = data['xx']
    yy = data['yy']
    Z = data['Z']

example_input = {
    'cap-shape': 'convex',
    'cap-surface': 'smooth',
    'cap-color': 'white',
    'bruises': 'bruises',
    'odor': 'pungent',
    'gill-attachment': 'free',
    'gill-spacing': 'close',
    'gill-size': 'narrow',
    'gill-color': 'black',
    'stalk-shape': 'enlarging',
    'stalk-root': 'equal',
    'stalk-surface-above-ring': 'smooth',
    'stalk-surface-below-ring': 'smooth',
    'stalk-color-above-ring': 'white',
    'stalk-color-below-ring': 'white',
    'veil-type': 'partial',
    'veil-color': 'white',
    'ring-number': 'one',
    'ring-type': 'pendant',
    'spore-print-color': 'black',
    'population': 'several',
    'habitat': 'grasses'
}

st.title('Mushroom Classification')

input_data = {}
keys = list(example_input.keys())
for i in range(0, len(keys), 4):
    cols = st.columns(4)
    for col, key in zip(cols, keys[i:i + 4]):
        options = list(loaded_dicts.get(key.replace('-', '_') + '_dict', {}).values())
        input_data[key] = col.selectbox(
            label=key.replace('_', ' ').title(),
            options=options,
            index=options.index(example_input[key]) if example_input[key] in options else 0,
        )

boundary_color1 = 'palegreen'
boundary_color2 = 'lightcoral'
custom_cmap = ListedColormap([boundary_color1, boundary_color2])

submitted = st.button('Submit')

if submitted:
    mushroom_input = np.array([list(input_data.values())])
    predict = input_predict(input_data, df_columns=df_columns, model_1=model_1, model_2=model_2, pca=pca_2d_loaded)
    print(str(predict[1][0]))
    if str(predict[1][0]) == 'poisonous':
        header_col = 'red'
    else:
        header_col = 'green'
    st.header(f':{header_col}[{str(predict[1][0]).upper()}]', anchor=False, divider=header_col)
    if not predict[2]:
        st.markdown(
            '<span style="color:red; font-weight:bold;">&#9888; Potential Misleading Plot</span>',
            unsafe_allow_html=True
        )
        st.markdown(
            '<span style="color:red; font-size:small;">The plotted data may be misleading because the Principal Component Analysis (PCA) used for dimensionality reduction only explains 30% of the variance in the original dataset.</span>',
            unsafe_allow_html=True
        )
    vec_2d_imput = predict[0]
    train_size = 50
    fig, ax = plt.subplots(figsize=(9, 7))
    ax.contourf(xx, yy, Z, np.linspace(0, 1, 3), alpha=0.3, cmap=custom_cmap)
    prediction_test = ax.scatter(vec_2d_imput[:, 0], vec_2d_imput[:, 1], marker='x', c='k', s=200,
                                 label=f'Prediction')
    train_edible_marker = ax.scatter([], [], c=boundary_color1, label='Edible', s=train_size, marker='o')
    train_poisonus_marker = ax.scatter([], [], c=boundary_color2, label='Poisonus', s=train_size, marker='o')
    ax.legend(handles=[train_edible_marker, train_poisonus_marker, prediction_test])
    ax.axis('off')
    plt.subplots_adjust(left=0, right=1, top=1, bottom=0)
    st.pyplot(fig)
    st.markdown(
        '<div style="text-align: center; font-style: italic; font-size:small;">- 2D plot of the model -</div>',
        unsafe_allow_html=True
    )

# streamlit run C:/Users/teoto/PycharmProjects/huggingface_app/app.py