predict.py

from load_data import *
from tensorflow.keras.models import load_model
import matplotlib.pyplot as plt
from matplotlib.style import use
use("seaborn")


def plot_predict_test_data(ground_truth_data, input_data, predictions):
    fig, ax = plt.subplots(figsize=(16, 9))
    plt.plot(ground_truth_data, color="tab:red", label="True Value")
    ax.plot(
        range(len(input_data) + 1, len(input_data) + 1 + len(predictions)),
        predictions,
        color="tab:blue",
        label="Predicted Testing Value"
    )
    plt.legend()
    plt.show()


def plot_all_data(input_data, predictions, fig, ax, **label):
    ax.plot(input_data, color="tab:red", label=label["ground_truth"])
    ax.plot(predictions, color="tab:blue", label=label["predict_value"])
    return fig, ax


def predict_func(input_data, model_name, scaler):
    model = load_model(model_name)
    predictions = model.predict(input_data)
    predictions = scaler.inverse_transform(predictions)
    return predictions


if __name__ == "__main__":
    csv_file = "TY_climate_2017_2018.csv"
    # column_name = "TT-Avg(℃)"  # column_name: TT-Avg(℃), MT-Avg(g)
    column_name = "MT-Avg(g)"

    # load date
    train_data, test_data = load_data(csv_file, column_name)
    train_data, scaler_train = data_preprocessing(train_data)
    test_data, scaler_test = data_preprocessing(test_data)

    # get ground truth
    df = pd.read_csv(csv_file)
    df = df[column_name].values
    df = df.reshape(-1, 1)

    # create data
    x_train, y_train = create_dataset(train_data)
    x_test, y_test = create_dataset(test_data)

    # reshape
    x_train = x_train.reshape(x_train.shape[0], 1, 1)
    y_train = y_train.reshape(y_train.shape[0], 1, 1)
    x_test = x_test.reshape(x_test.shape[0], 1, 1)
    y_test = y_test.reshape(y_test.shape[0], 1, 1)

    # predict all data
    all_data, scaler_all_data = data_preprocessing(df)
    all_data_x, all_data_y = create_dataset(all_data)
    all_data_x = all_data_x.reshape(all_data_x.shape[0], 1, 1)
    all_data_y = all_data_y.reshape(all_data_y.shape[0], 1, 1)

    # ---------------------------------------------------------------------------
    # ------------------------predict test part----------------------------------
    # model_name = "saved_models_tt_avg/LSTM_002.h5"
    # predictions = predict_func(x_test, model_name, scaler_test)

    # model_name = "saved_models_mt_avg/LSTM_002.h5"
    # predictions = predict_func(all_data_x, model_name, scaler_all_data)

    # ------------------------predict all data-----------------------------------
    # model_name = "saved_models_tt_avg/LSTM_002.h5"
    # predictions = predict_func(all_data_x, model_name, scaler_all_data)

    model_name = "saved_models_mt_avg/LSTM_002.h5"
    predictions = predict_func(all_data_x, model_name, scaler_all_data)
    # ---------------------------------------------------------------------------

    # plot predicted part and all ground truth
    # plot_predict_test_data(df, y_train, predictions)

    # plot all ground truth data and all predicted data
    labels = {
        "ground_truth": "True Testing Value",
        "predict_value": "Predicted Testing Value"
    }
    fig, ax = plt.subplots(figsize=(16, 9))
    fig, ax = plot_all_data(df[1:], predictions, fig, ax, **labels)
    plt.legend()
    plt.show()