weather_cnn_comparison.py

from keras.models import Sequential
from keras.layers import BatchNormalization, Conv2D, Conv2DTranspose
from keras.optimizers import Adam
import matplotlib.pyplot as plt
import numpy as np

def GetModel():
    """ Encoder-Decoder Approach """
    """
    model = Sequential()
    model.add(BatchNormalization(axis=3, input_shape=(80, 120, 3)))
    model.add(Conv2D(64, (5, 5), strides=(2, 2), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2D(128, (3, 3), strides=(2, 2), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2D(256, (3, 3), strides=(2, 2), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2DTranspose(128, (3, 3), strides=(2, 2), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2DTranspose(64, (3, 3), strides=(2, 2), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2DTranspose(1, (5, 5), strides=(2, 2), activation='relu', padding='same'))
    """

    """ Constant size convolution Approach """
    model = Sequential()
    model.add(BatchNormalization(axis=3, input_shape=(80, 120, 3)))
    model.add(Conv2D(64, (5, 5), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2D(128, (5, 5), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2D(256, (5, 5), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2D(128, (5, 5), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2D(64, (5, 5), activation='relu', padding='same'))
    model.add(BatchNormalization(axis=3))
    model.add(Conv2D(1, (5, 5), activation='relu', padding='same'))

    #model.compile(loss='binary_crossentropy', optimizer=Adam(lr=0.001), metrics=['accuracy'])
    #model.compile(loss='binary_crossentropy', optimizer=Adam(lr=0.001), metrics=['mse'])
    model.compile(loss='mean_absolute_error', optimizer=Adam(lr=0.001), metrics=['mae'])

    print(model.summary())

    return model

x = np.load("3zlevels.npy")
y = 1000*np.expand_dims(np.load("full_tp_1980_2016.npy"), axis=3)#[44000:, :], axis=3)

print("rain mean", y.mean())
print(x.shape)
print(y.shape)

x_train = x[:4000, :, :, :]
y_train = y[:4000, :]

x_test = x[4000:5000, :, :, :]
y_test = y[4000:5000, :]
print("train x mean", x[:4000, :].mean())
print("test x mean", x[4000:5000, :].mean())

model = GetModel()
history = model.fit(x_train, y_train, epochs=30, verbose=1, validation_data=(x_test, y_test))