RegressionMultiInput.py

import numpy as np
import pandas as pd
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import BatchNormalization, Conv2D, MaxPooling2D, Conv2DTranspose, Reshape
from tensorflow.keras.layers import Activation, Dropout, Dense, Flatten, Input, UpSampling2D, Concatenate
from tensorflow.keras.models import Model
from tensorflow.keras.models import load_model
from tensorflow.keras.optimizers import Adam
from sklearn.model_selection import train_test_split
import tensorflow as tf
import pickle5 as pickle
from tensorflow.keras import backend as K
import locale

def load_dataset(path="/content/drive/MyDrive/DeepLearning/DataEmoda.npy"):
    with open(path, "rb") as fh:
      df = pickle.load(fh)
    return df

def create_input_output(df, input_labels, output_labels):
    inputImage = []
    outputImage = []
    for i in df.index:
        inputImage.append(df[input_labels][i].reshape(256,256,1))
        outputImage.append(df[output_labels][i].reshape(256,256,1))
    return np.array(inputImage), np.array(outputImage)

def create_input(df, input_labels):
    return df[input_labels].values

def create_mlp(dim):
    model = Sequential()
    model.add(Dense(64, input_dim=dim, activation='relu'))
    model.add(Dense(1024, activation='relu'))
    model.add(Dense(256*256, activation='relu'))
    model.build((None, 256*256))

    return model

def create_cnn(width, height, depth):
    inputShape = (height, width, depth)
    
    #Main Branch
    inputs = Input(shape=inputShape)
    x = Conv2D(64, (3,3), padding="same")(inputs)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    ax = MaxPooling2D(pool_size=(2,2))(x)

    #Branch 1
    x = Conv2D(32, (3,3), padding="same")(ax)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    bx = MaxPooling2D(pool_size=(2,2))(x)

    #Branch 2
    x = Conv2D(32, (3,3), padding="same")(bx)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    cx = MaxPooling2D(pool_size=(2,2))(x)

    #Branch 3
    x = Conv2D(32, (3,3), padding="same")(cx)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    x = MaxPooling2D(pool_size=(2,2))(x)
    x = Conv2D(32, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    x = UpSampling2D(size=(2,2))(x)
    x = Conv2DTranspose(32, (3,3), padding="same")(x)

    #Branch 2
    x = Concatenate()([x, cx])
    x = Conv2D(32, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    x = UpSampling2D(size=(2,2))(x)
    x = Conv2DTranspose(32, (3,3), padding="same")(x)

    #Branch 1
    x = Concatenate()([x, bx])
    x = Conv2D(32, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    x = UpSampling2D(size=(2,2))(x)
    x = Conv2DTranspose(32, (3,3), padding="same")(x)
    
    #Main Branch
    x = Concatenate()([x,ax])
    x = Conv2D(16, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    
    model = Model(inputs, x)
    
    return model
    
def full_model(cnn_model, mlp_model):
    
    x = cnn_model.output
    cx = mlp_model.output
    
    conv_shape = K.int_shape(x)
    
    cx = Reshape((conv_shape[1],conv_shape[2],int(conv_shape[3]/4)))(cx)
    
    x = Concatenate()([x,cx])
    
    x = Conv2D(16, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    
    x = Conv2D(16, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    
    x = Conv2D(16, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    
    x = Conv2D(16, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    
    x = Conv2D(16, (3,3), padding="same")(x)
    x = Activation("relu")(x)
    x = BatchNormalization()(x)
    
    x = UpSampling2D(size=(2,2))(x)
    x = Conv2DTranspose(1, (3,3), padding="same", activation="linear")(x)
    
    model = Model(inputs=[cnn_model.input, mlp_model.input], outputs = x)
    
    return model   


df = load_dataset()
(inputImages,outputImages) = create_input_output(df, 'bathy', 'hs')
(inputImages, outputImages) = ((inputImages - np.nanmean(inputImages))/np.nanstd(inputImages), (outputImages - np.nanmean(outputImages)/np.nanstd(outputImages)))

inputAttr = create_input(df, ['$\eta$', '$\zeta$','$\theta_{wave}$'])
inputAttr[:,0] = (inputAttr[:,0] - np.mean(inputAttr[:,0])) / np.std(inputAttr[:,0])
inputAttr[:,1] = (inputAttr[:,1] - np.mean(inputAttr[:,1])) / np.std(inputAttr[:,1])
inputAttr[:,2] = inputAttr[:,2] / (2*np.pi)

(inputImages, outputImages) = (np.nan_to_num(inputImages,nan=-10.), np.nan_to_num(outputImages, nan=-10.))

split = train_test_split(inputImages, inputAttr, outputImages, test_size=0.25, random_state=42)
(trainImgX, testImgX, trainAttrX, testAttrX, trainY, testY) = split

cnn_model = create_cnn(256, 256, 1, regress=True)
mlp_model = create_mlp(trainAttrX.shape[1])

model = full_model(cnn_model, mlp_model)
opt = Adam(learning_rate=1e-4, decay=1e-4/200)
model.compile(loss="mean_squared_error", optimizer=opt)

print("[INFO] training model...")
model.fit(x=[trainImgX, trainAttrX], y=trainY,
          validation_data=([testImgX, testAttrX], testY),
          epochs=200, batch_size=8)

model.save('/content/drive/MyDrive/DeepLearning/ModelV6')



#%% Plotting
from tensorflow.keras.models import load_model
import matplotlib.pyplot as plt
model = load_model('/content/drive/MyDrive/DeepLearning/ModelV6', compile = True)

Prediction = model.predict([testImgX[0:7], testAttrX[0:7]])[6][:,:,0]
Truehs = testY[6][:,:,0]

fig = plt.figure(figsize=(6,3))

ax = fig.add_subplot(1,2,1)
ax.set_title('colorMap')
plt.imshow(Prediction)

qx = fig.add_subplot(1,2,2)
plt.imshow(Truehs)

cax = fig.add_axes([0.12, 0.1, 0.78, 0.8])
cax.get_xaxis().set_visible(False)
cax.get_yaxis().set_visible(False)
cax.patch.set_alpha(0)
cax.set_frame_on(False)
plt.colorbar(orientation='vertical')
plt.show()