Tensorflow sequential API

A Sequential model is appropriate for a plain stack of layers where each layer has exactly one input tensor and one output tensor.

Get an overview on keras here.

import tensorflow as tf
from tensorflow import keras
import os
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

fashion_mnist_data = keras.datasets.fashion_mnist
(all_x_train, all_y_train), (x_test, y_test) = fashion_mnist_data.load_data()

all_x_train = all_x_train.astype('float32')
x_test = x_test.astype('float32')


print(f"all_x_train.shape = {all_x_train.shape}")
print(f"all_x_train[0].shape = {all_x_train[0].shape}")
print(f"all_x_train[0].dtype = {all_x_train[0].dtype}")



x_validation, x_train = all_x_train[:5000] / 255.0, all_x_train[5000:] / 255.0
y_validation, y_train = all_y_train[:5000], all_y_train[5000:]

print(f"x_train.shape = {x_train.shape}")
print(f"x_train[0].shape = {x_train[0].shape}")
print(f"x_train[0].dtype = {x_train[0].dtype}")



fashion_mnist_class_names = ["T-shirt/top", "Trouser", "Pullover", "Dress", "Coat", "Sandal", "Shirt", "Sneaker", "Bag", "Ankle boot"]


for cls in range(10):
    print(cls, ":",fashion_mnist_class_names[y_train[cls]])


for i in range(5):
    my_img= x_train[i]
    my_img_class = y_train[i]
    my_img_class_name = fashion_mnist_class_names[my_img_class]
    plt.imshow(my_img)
    plt.title(my_img_class_name)
    plt.show()

all_x_train.shape = (60000, 28, 28)
all_x_train[0].shape = (28, 28)
all_x_train[0].dtype = float32
x_train.shape = (55000, 28, 28)
x_train[0].shape = (28, 28)
x_train[0].dtype = float32
0 : Coat
1 : T-shirt/top
2 : Sneaker
3 : Ankle boot
4 : Ankle boot
5 : Ankle boot
6 : Coat
7 : Coat
8 : Dress
9 : Coat

png

model = keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28, 28]))
model.add(keras.layers.Dense(300, activation="relu"))
model.add(keras.layers.Dense(150, activation="relu"))
model.add(keras.layers.Dense(10, activation="softmax"))

model.summary()


print(model.layers)
print(model.layers[2].name)

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 flatten (Flatten)           (None, 784)               0

 dense (Dense)               (None, 300)               235500

 dense_1 (Dense)             (None, 150)               45150

 dense_2 (Dense)             (None, 10)                1510

=================================================================
Total params: 282160 (1.08 MB)
Trainable params: 282160 (1.08 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________
[<keras.src.layers.reshaping.flatten.Flatten object at 0x7a4efec1a320>, <keras.src.layers.core.dense.Dense object at 0x7a4efc795690>, <keras.src.layers.core.dense.Dense object at 0x7a4efc794220>, <keras.src.layers.core.dense.Dense object at 0x7a4efc794190>]
dense_1

Compilation

model.compile(loss="sparse_categorical_crossentropy", optimizer="sgd", metrics=["accuracy"])

Training

results = model.fit(x_train, y_train, epochs=10,\
                    validation_data=(x_validation, y_validation))

Epoch 1/10
1719/1719 [==============================] - 20s 11ms/step - loss: 0.7264 - accuracy: 0.7619 - val_loss: 0.5066 - val_accuracy: 0.8276
Epoch 2/10
1719/1719 [==============================] - 16s 9ms/step - loss: 0.4924 - accuracy: 0.8283 - val_loss: 0.4655 - val_accuracy: 0.8408
Epoch 3/10
1719/1719 [==============================] - 16s 10ms/step - loss: 0.4453 - accuracy: 0.8443 - val_loss: 0.4157 - val_accuracy: 0.8594
Epoch 4/10
1719/1719 [==============================] - 10s 6ms/step - loss: 0.4173 - accuracy: 0.8536 - val_loss: 0.4018 - val_accuracy: 0.8642
Epoch 5/10
1719/1719 [==============================] - 9s 5ms/step - loss: 0.3964 - accuracy: 0.8609 - val_loss: 0.3930 - val_accuracy: 0.8650
Epoch 6/10
1719/1719 [==============================] - 10s 6ms/step - loss: 0.3798 - accuracy: 0.8657 - val_loss: 0.3848 - val_accuracy: 0.8658
Epoch 7/10
1719/1719 [==============================] - 10s 6ms/step - loss: 0.3658 - accuracy: 0.8706 - val_loss: 0.3583 - val_accuracy: 0.8784
Epoch 8/10
1719/1719 [==============================] - 10s 6ms/step - loss: 0.3544 - accuracy: 0.8748 - val_loss: 0.3682 - val_accuracy: 0.8728
Epoch 9/10
1719/1719 [==============================] - 10s 6ms/step - loss: 0.3432 - accuracy: 0.8782 - val_loss: 0.3495 - val_accuracy: 0.8772
Epoch 10/10
1719/1719 [==============================] - 11s 7ms/step - loss: 0.3347 - accuracy: 0.8805 - val_loss: 0.3383 - val_accuracy: 0.8824

Results Analysis

pd.DataFrame(results.history).plot(figsize=(8, 5))
plt.grid(True)
plt.gca().set_ylim(0, 1)
plt.show()

png

Model evaluation

x_test= x_test/255.0
print("Evaluation du modèle :")
model.evaluate(x_test, y_test)

Evaluation du modèle :
313/313 [==============================] - 2s 5ms/step - loss: 0.3756 - accuracy: 0.8681





[0.3756197392940521, 0.8680999875068665]

Model prediction

x_new = x_test[11:20]
y_prob = model.predict(x_new)
print(f"y_prob = {y_prob.round(2)}")
y_prediction = model.predict(x_new)
print(f"y_prediction = {y_prediction}")

1/1 [==============================] - 0s 26ms/step
y_prob = [[0.   0.   0.   0.   0.   0.94 0.   0.03 0.   0.03]
 [0.   0.   0.   0.01 0.   0.37 0.   0.45 0.17 0.  ]
 [0.   0.01 0.   0.99 0.   0.   0.   0.   0.   0.  ]
 [0.   0.   0.2  0.   0.75 0.   0.02 0.   0.02 0.  ]
 [0.   0.99 0.   0.   0.   0.   0.   0.   0.   0.  ]
 [0.02 0.   0.96 0.   0.01 0.   0.01 0.   0.   0.  ]
 [0.   0.   0.78 0.   0.16 0.   0.06 0.   0.   0.  ]
 [0.   0.   0.   0.   0.   0.   0.   0.   1.   0.  ]
 [0.95 0.   0.   0.   0.   0.   0.05 0.   0.   0.  ]]
1/1 [==============================] - 0s 26ms/step
y_prediction = [[3.1884041e-04 9.0832909e-06 2.2825778e-04 7.7158285e-05 2.8249460e-05
  9.4131207e-01 4.9991539e-04 2.9197719e-02 3.3048738e-03 2.5023883e-02]
 [2.8612296e-04 2.5517552e-04 5.2534125e-04 1.1631102e-02 5.4271437e-05
  3.6781955e-01 3.3234953e-04 4.5135418e-01 1.6645952e-01 1.2823747e-03]
 [1.8247330e-04 1.0255000e-02 1.5272074e-04 9.8864001e-01 1.5142665e-04
  5.2109713e-06 5.4259861e-05 1.3752261e-05 5.3833018e-04 6.7150231e-06]
 [4.9151644e-05 6.2005820e-05 2.0411910e-01 3.4922175e-04 7.4825048e-01
  1.1576705e-09 2.2210740e-02 1.0316979e-09 2.4959395e-02 9.1224861e-10]
 [5.2958861e-04 9.9382234e-01 1.3185898e-04 4.9118144e-03 5.0863827e-04
  1.1390349e-06 6.2134888e-05 2.0139789e-06 2.8816045e-05 1.5881855e-06]
 [1.7020402e-02 1.6831007e-03 9.6144742e-01 9.4521814e-04 5.4272199e-03
  4.9377145e-06 1.3044720e-02 1.3532654e-06 4.2115254e-04 4.3969744e-06]
 [1.0709631e-03 5.9443355e-05 7.8045821e-01 2.2999026e-04 1.5594642e-01
  9.6357153e-09 6.1186492e-02 3.0131968e-08 1.0483154e-03 8.8528068e-08]
 [9.1176917e-05 1.4379219e-06 1.7904839e-04 2.8298934e-06 5.8932448e-05
  9.7865406e-05 1.8255299e-04 1.4076787e-05 9.9936873e-01 3.3107310e-06]
 [9.4995093e-01 1.5609858e-05 3.6273990e-04 7.0983241e-04 1.8508168e-06
  1.7300086e-09 4.8956469e-02 1.3470089e-08 2.5965862e-06 3.4715937e-09]]

A Sequential model is not appropriate when:

Your model has multiple inputs or multiple outputs
Any of your layers has multiple inputs or multiple outputs
You need to do layer sharing
You want non-linear topology (e.g. a residual connection, a multi-branch model)

Tensorflow functional API

The Keras functional API is a way to create models that are more flexible than the keras.Sequential API. The functional API can handle models with non-linear topology, shared layers, and even multiple inputs or outputs.


from sklearn.datasets import fetch_california_housing
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

housing = fetch_california_housing()
all_x_train, x_test, all_y_train, y_test = train_test_split(housing.data, housing.target)


x_train, x_validation, y_train, y_validation = train_test_split(all_x_train, all_y_train)
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
x_validation_scaled = scaler.transform(x_validation)
x_test_scaled = scaler.transform(x_test)

input = keras.layers.Input(shape=x_train_scaled.shape[1:])
hidden1 = keras.layers.Dense(30, activation="relu")(input)
hidden2 = keras.layers.Dense(30, activation="relu")(hidden1)
hidden3 = keras.layers.Dense(30, activation="relu")(hidden2)
output = keras.layers.Dense(1)(hidden3)

model = keras.models.Model(inputs=[input], outputs=[output])

model.compile(loss="mse", optimizer="rmsprop", metrics=["mae"])
ressults = model.fit(x_train_scaled, y_train, epochs=40, validation_data=(x_validation_scaled, y_validation))

res_eval = model.evaluate(x_test_scaled, y_test)

Epoch 1/40
363/363 [==============================] - 2s 3ms/step - loss: 1.3137 - mae: 0.7744 - val_loss: 0.4826 - val_mae: 0.5053
Epoch 2/40
363/363 [==============================] - 1s 4ms/step - loss: 0.4395 - mae: 0.4717 - val_loss: 0.3986 - val_mae: 0.4432
Epoch 3/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3878 - mae: 0.4435 - val_loss: 0.3695 - val_mae: 0.4184
Epoch 4/40
363/363 [==============================] - 2s 6ms/step - loss: 0.3679 - mae: 0.4315 - val_loss: 0.3603 - val_mae: 0.4186
Epoch 5/40
363/363 [==============================] - 2s 4ms/step - loss: 0.3657 - mae: 0.4233 - val_loss: 0.3431 - val_mae: 0.4132
Epoch 6/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3511 - mae: 0.4183 - val_loss: 0.3349 - val_mae: 0.4039
Epoch 7/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3474 - mae: 0.4127 - val_loss: 0.3276 - val_mae: 0.4054
Epoch 8/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3348 - mae: 0.4080 - val_loss: 0.3299 - val_mae: 0.3922
Epoch 9/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3313 - mae: 0.4019 - val_loss: 0.3175 - val_mae: 0.3899
Epoch 10/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3208 - mae: 0.3983 - val_loss: 0.3185 - val_mae: 0.3891
Epoch 11/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3256 - mae: 0.3943 - val_loss: 0.3218 - val_mae: 0.4053
Epoch 12/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3161 - mae: 0.3885 - val_loss: 0.3141 - val_mae: 0.3909
Epoch 13/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3054 - mae: 0.3863 - val_loss: 0.3059 - val_mae: 0.3810
Epoch 14/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3020 - mae: 0.3830 - val_loss: 0.3081 - val_mae: 0.3870
Epoch 15/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2982 - mae: 0.3805 - val_loss: 0.2988 - val_mae: 0.3683
Epoch 16/40
363/363 [==============================] - 1s 4ms/step - loss: 0.3020 - mae: 0.3787 - val_loss: 0.3170 - val_mae: 0.3918
Epoch 17/40
363/363 [==============================] - 2s 4ms/step - loss: 0.2956 - mae: 0.3756 - val_loss: 0.3040 - val_mae: 0.3735
Epoch 18/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2984 - mae: 0.3753 - val_loss: 0.3033 - val_mae: 0.3791
Epoch 19/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2913 - mae: 0.3721 - val_loss: 0.3025 - val_mae: 0.3765
Epoch 20/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2875 - mae: 0.3716 - val_loss: 0.3015 - val_mae: 0.3717
Epoch 21/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2911 - mae: 0.3690 - val_loss: 0.3252 - val_mae: 0.4026
Epoch 22/40
363/363 [==============================] - 1s 3ms/step - loss: 0.3021 - mae: 0.3701 - val_loss: 0.3102 - val_mae: 0.3670
Epoch 23/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2868 - mae: 0.3675 - val_loss: 0.2921 - val_mae: 0.3669
Epoch 24/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2871 - mae: 0.3663 - val_loss: 0.2926 - val_mae: 0.3615
Epoch 25/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2849 - mae: 0.3647 - val_loss: 0.2992 - val_mae: 0.3779
Epoch 26/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2785 - mae: 0.3638 - val_loss: 0.2973 - val_mae: 0.3587
Epoch 27/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2770 - mae: 0.3627 - val_loss: 0.2911 - val_mae: 0.3624
Epoch 28/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2839 - mae: 0.3622 - val_loss: 0.2907 - val_mae: 0.3567
Epoch 29/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2766 - mae: 0.3621 - val_loss: 0.2958 - val_mae: 0.3631
Epoch 30/40
363/363 [==============================] - 2s 4ms/step - loss: 0.2765 - mae: 0.3597 - val_loss: 0.2879 - val_mae: 0.3657
Epoch 31/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2758 - mae: 0.3598 - val_loss: 0.2945 - val_mae: 0.3584
Epoch 32/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2722 - mae: 0.3580 - val_loss: 0.2875 - val_mae: 0.3611
Epoch 33/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2824 - mae: 0.3592 - val_loss: 0.2927 - val_mae: 0.3670
Epoch 34/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2741 - mae: 0.3578 - val_loss: 0.2839 - val_mae: 0.3552
Epoch 35/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2707 - mae: 0.3559 - val_loss: 0.2866 - val_mae: 0.3581
Epoch 36/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2732 - mae: 0.3556 - val_loss: 0.2865 - val_mae: 0.3572
Epoch 37/40
363/363 [==============================] - 1s 2ms/step - loss: 0.2692 - mae: 0.3545 - val_loss: 0.3431 - val_mae: 0.4224
Epoch 38/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2700 - mae: 0.3545 - val_loss: 0.2892 - val_mae: 0.3603
Epoch 39/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2682 - mae: 0.3532 - val_loss: 0.3023 - val_mae: 0.3820
Epoch 40/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2665 - mae: 0.3527 - val_loss: 0.2859 - val_mae: 0.3591
162/162 [==============================] - 0s 3ms/step - loss: 0.2842 - mae: 0.3563

pd.DataFrame(ressults.history).plot(figsize=(8, 5))
plt.grid(True)
plt.gca().set_ylim(0, 1)
plt.show()

png

Tensorflow subclassing API

housing = fetch_california_housing()
all_x_train, x_test, all_y_train, y_test = train_test_split(housing.data, housing.target)


x_train, x_validation, y_train, y_validation = train_test_split(all_x_train, all_y_train)
scaler = StandardScaler()
x_train_scaled = scaler.fit_transform(x_train)
x_validation_scaled = scaler.transform(x_validation)
x_test_scaled = scaler.transform(x_test)


class Exemple_SubclassingAPI(keras.models.Model):

    def __init__(self, nb_unit_1, nb_unit_2, activation="relu", **kwargs):
        super().__init__(**kwargs)
        self.hidden1 = keras.layers.Dense(nb_unit_1, activation=activation)
        self.hidden2 = keras.layers.Dense(nb_unit_2, activation=activation)
        self.res = keras.layers.Dense(1)


    def call(self, inputs):
        input = inputs
        hidden1 = self.hidden1(input)
        hidden2 = self.hidden2(hidden1)
        res = self.res(hidden2)
        return res

model = Exemple_SubclassingAPI(30,15)

model.compile(loss="mse", optimizer="rmsprop", metrics=["mae"])

resssults = model.fit(x_train_scaled, y_train, epochs=40, validation_data=(x_validation_scaled, y_validation))

mse_test = model.evaluate(x_test_scaled, y_test)

Epoch 1/40
363/363 [==============================] - 2s 3ms/step - loss: 0.2794 - mae: 0.3631 - val_loss: 0.2949 - val_mae: 0.3804
Epoch 2/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2802 - mae: 0.3619 - val_loss: 0.2879 - val_mae: 0.3621
Epoch 3/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2789 - mae: 0.3617 - val_loss: 0.2895 - val_mae: 0.3716
Epoch 4/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2767 - mae: 0.3618 - val_loss: 0.2884 - val_mae: 0.3704
Epoch 5/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2750 - mae: 0.3608 - val_loss: 0.2977 - val_mae: 0.3773
Epoch 6/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2747 - mae: 0.3606 - val_loss: 0.2963 - val_mae: 0.3789
Epoch 7/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2793 - mae: 0.3604 - val_loss: 0.2837 - val_mae: 0.3598
Epoch 8/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2794 - mae: 0.3597 - val_loss: 0.2897 - val_mae: 0.3691
Epoch 9/40
363/363 [==============================] - 1s 2ms/step - loss: 0.2744 - mae: 0.3583 - val_loss: 0.2899 - val_mae: 0.3746
Epoch 10/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2739 - mae: 0.3586 - val_loss: 0.2890 - val_mae: 0.3580
Epoch 11/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2716 - mae: 0.3579 - val_loss: 0.2855 - val_mae: 0.3627
Epoch 12/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2717 - mae: 0.3586 - val_loss: 0.2929 - val_mae: 0.3627
Epoch 13/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2762 - mae: 0.3579 - val_loss: 0.2827 - val_mae: 0.3598
Epoch 14/40
363/363 [==============================] - 1s 2ms/step - loss: 0.2740 - mae: 0.3569 - val_loss: 0.2930 - val_mae: 0.3733
Epoch 15/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2799 - mae: 0.3571 - val_loss: 0.2817 - val_mae: 0.3625
Epoch 16/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2736 - mae: 0.3558 - val_loss: 0.2816 - val_mae: 0.3646
Epoch 17/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2716 - mae: 0.3557 - val_loss: 0.2841 - val_mae: 0.3579
Epoch 18/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2678 - mae: 0.3556 - val_loss: 0.2857 - val_mae: 0.3669
Epoch 19/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2746 - mae: 0.3558 - val_loss: 0.2868 - val_mae: 0.3591
Epoch 20/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2701 - mae: 0.3551 - val_loss: 0.2845 - val_mae: 0.3663
Epoch 21/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2674 - mae: 0.3540 - val_loss: 0.2860 - val_mae: 0.3556
Epoch 22/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2682 - mae: 0.3540 - val_loss: 0.2826 - val_mae: 0.3598
Epoch 23/40
363/363 [==============================] - 1s 2ms/step - loss: 0.2743 - mae: 0.3547 - val_loss: 0.2880 - val_mae: 0.3585
Epoch 24/40
363/363 [==============================] - 1s 2ms/step - loss: 0.2682 - mae: 0.3544 - val_loss: 0.2838 - val_mae: 0.3671
Epoch 25/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2713 - mae: 0.3551 - val_loss: 0.2789 - val_mae: 0.3591
Epoch 26/40
363/363 [==============================] - 1s 2ms/step - loss: 0.2659 - mae: 0.3532 - val_loss: 0.2781 - val_mae: 0.3566
Epoch 27/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2654 - mae: 0.3531 - val_loss: 0.2812 - val_mae: 0.3605
Epoch 28/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2679 - mae: 0.3519 - val_loss: 0.2786 - val_mae: 0.3603
Epoch 29/40
363/363 [==============================] - 2s 5ms/step - loss: 0.2653 - mae: 0.3529 - val_loss: 0.2821 - val_mae: 0.3586
Epoch 30/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2672 - mae: 0.3528 - val_loss: 0.2844 - val_mae: 0.3587
Epoch 31/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2652 - mae: 0.3521 - val_loss: 0.2798 - val_mae: 0.3528
Epoch 32/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2643 - mae: 0.3513 - val_loss: 0.2789 - val_mae: 0.3564
Epoch 33/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2629 - mae: 0.3501 - val_loss: 0.2818 - val_mae: 0.3669
Epoch 34/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2630 - mae: 0.3504 - val_loss: 0.2914 - val_mae: 0.3762
Epoch 35/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2614 - mae: 0.3499 - val_loss: 0.2879 - val_mae: 0.3545
Epoch 36/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2653 - mae: 0.3500 - val_loss: 0.2787 - val_mae: 0.3591
Epoch 37/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2625 - mae: 0.3501 - val_loss: 0.2844 - val_mae: 0.3694
Epoch 38/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2625 - mae: 0.3504 - val_loss: 0.2844 - val_mae: 0.3620
Epoch 39/40
363/363 [==============================] - 1s 3ms/step - loss: 0.2618 - mae: 0.3510 - val_loss: 0.2775 - val_mae: 0.3585
Epoch 40/40
363/363 [==============================] - 1s 4ms/step - loss: 0.2614 - mae: 0.3505 - val_loss: 0.2790 - val_mae: 0.3635
162/162 [==============================] - 0s 2ms/step - loss: 0.2890 - mae: 0.3684

pd.DataFrame(resssults.history).plot(figsize=(8, 5))
plt.grid(True)
plt.gca().set_ylim(0, 1)
plt.show()

png

Let's use the sequential API for regression

model = keras.models.Sequential()
model.add(keras.layers.Dense(30, activation="relu", input_shape=x_train_scaled.shape[1:]))
model.add(keras.layers.Dense(15, activation="relu"))
model.add(keras.layers.Dense(8, activation="relu"))
model.add(keras.layers.Dense(1))



model.summary()

Model: "sequential_1"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 dense_10 (Dense)            (None, 30)                270

 dense_11 (Dense)            (None, 15)                465

 dense_12 (Dense)            (None, 8)                 128

 dense_13 (Dense)            (None, 1)                 9

=================================================================
Total params: 872 (3.41 KB)
Trainable params: 872 (3.41 KB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

model.compile(loss="mse", optimizer="rmsprop", metrics=["mae"])
ressults = model.fit(x_train_scaled, y_train, epochs=100, \
                     validation_data=(x_validation_scaled, y_validation))

mae_test = model.evaluate(x_test_scaled, y_test)

Epoch 1/100
363/363 [==============================] - 4s 8ms/step - loss: 1.4869 - mae: 0.7407 - val_loss: 0.4752 - val_mae: 0.5042
Epoch 2/100
363/363 [==============================] - 2s 5ms/step - loss: 0.4607 - mae: 0.4825 - val_loss: 0.4075 - val_mae: 0.4410
Epoch 3/100
363/363 [==============================] - 2s 4ms/step - loss: 0.4037 - mae: 0.4547 - val_loss: 0.3763 - val_mae: 0.4241
Epoch 4/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3820 - mae: 0.4393 - val_loss: 0.3562 - val_mae: 0.4234
Epoch 5/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3829 - mae: 0.4293 - val_loss: 0.3478 - val_mae: 0.4255
Epoch 6/100
363/363 [==============================] - 2s 4ms/step - loss: 0.3530 - mae: 0.4213 - val_loss: 0.3571 - val_mae: 0.4422
Epoch 7/100
363/363 [==============================] - 2s 4ms/step - loss: 0.3416 - mae: 0.4163 - val_loss: 0.3383 - val_mae: 0.4139
Epoch 8/100
363/363 [==============================] - 2s 4ms/step - loss: 0.3454 - mae: 0.4132 - val_loss: 0.3222 - val_mae: 0.3967
Epoch 9/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3344 - mae: 0.4061 - val_loss: 0.3199 - val_mae: 0.3984
Epoch 10/100
363/363 [==============================] - 1s 4ms/step - loss: 0.3286 - mae: 0.4024 - val_loss: 0.3219 - val_mae: 0.3981
Epoch 11/100
363/363 [==============================] - 2s 6ms/step - loss: 0.3222 - mae: 0.3995 - val_loss: 0.3233 - val_mae: 0.3966
Epoch 12/100
363/363 [==============================] - 2s 4ms/step - loss: 0.3138 - mae: 0.3962 - val_loss: 0.3120 - val_mae: 0.3947
Epoch 13/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3181 - mae: 0.3943 - val_loss: 0.3181 - val_mae: 0.3995
Epoch 14/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3221 - mae: 0.3921 - val_loss: 0.3130 - val_mae: 0.3941
Epoch 15/100
363/363 [==============================] - 1s 4ms/step - loss: 0.3204 - mae: 0.3887 - val_loss: 0.3254 - val_mae: 0.4120
Epoch 16/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3186 - mae: 0.3879 - val_loss: 0.3186 - val_mae: 0.3990
Epoch 17/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3158 - mae: 0.3866 - val_loss: 0.3057 - val_mae: 0.3842
Epoch 18/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3244 - mae: 0.3845 - val_loss: 0.3099 - val_mae: 0.3735
Epoch 19/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2990 - mae: 0.3817 - val_loss: 0.3139 - val_mae: 0.3884
Epoch 20/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3195 - mae: 0.3822 - val_loss: 0.3035 - val_mae: 0.3810
Epoch 21/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3014 - mae: 0.3815 - val_loss: 0.3034 - val_mae: 0.3782
Epoch 22/100
363/363 [==============================] - 1s 4ms/step - loss: 0.3015 - mae: 0.3795 - val_loss: 0.3002 - val_mae: 0.3767
Epoch 23/100
363/363 [==============================] - 2s 4ms/step - loss: 0.2970 - mae: 0.3775 - val_loss: 0.3005 - val_mae: 0.3860
Epoch 24/100
363/363 [==============================] - 2s 4ms/step - loss: 0.3021 - mae: 0.3780 - val_loss: 0.3064 - val_mae: 0.3802
Epoch 25/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2982 - mae: 0.3767 - val_loss: 0.3019 - val_mae: 0.3757
Epoch 26/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3018 - mae: 0.3764 - val_loss: 0.3202 - val_mae: 0.4158
Epoch 27/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3032 - mae: 0.3767 - val_loss: 0.3064 - val_mae: 0.3766
Epoch 28/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2974 - mae: 0.3736 - val_loss: 0.3231 - val_mae: 0.3856
Epoch 29/100
363/363 [==============================] - 2s 5ms/step - loss: 0.2889 - mae: 0.3739 - val_loss: 0.3226 - val_mae: 0.4116
Epoch 30/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2949 - mae: 0.3728 - val_loss: 0.2985 - val_mae: 0.3756
Epoch 31/100
363/363 [==============================] - 1s 4ms/step - loss: 0.3200 - mae: 0.3735 - val_loss: 0.3073 - val_mae: 0.3723
Epoch 32/100
363/363 [==============================] - 1s 4ms/step - loss: 0.3222 - mae: 0.3738 - val_loss: 0.3109 - val_mae: 0.3798
Epoch 33/100
363/363 [==============================] - 2s 5ms/step - loss: 0.3014 - mae: 0.3716 - val_loss: 0.3142 - val_mae: 0.4042
Epoch 34/100
363/363 [==============================] - 2s 6ms/step - loss: 0.2876 - mae: 0.3695 - val_loss: 0.3123 - val_mae: 0.3877
Epoch 35/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3072 - mae: 0.3703 - val_loss: 0.3008 - val_mae: 0.3816
Epoch 36/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2996 - mae: 0.3690 - val_loss: 0.2944 - val_mae: 0.3756
Epoch 37/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2945 - mae: 0.3690 - val_loss: 0.2946 - val_mae: 0.3755
Epoch 38/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3074 - mae: 0.3690 - val_loss: 0.3004 - val_mae: 0.3707
Epoch 39/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3214 - mae: 0.3692 - val_loss: 0.3042 - val_mae: 0.3887
Epoch 40/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2964 - mae: 0.3680 - val_loss: 0.3038 - val_mae: 0.3858
Epoch 41/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2865 - mae: 0.3680 - val_loss: 0.3029 - val_mae: 0.3732
Epoch 42/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3014 - mae: 0.3679 - val_loss: 0.2927 - val_mae: 0.3762
Epoch 43/100
363/363 [==============================] - 1s 4ms/step - loss: 0.3001 - mae: 0.3675 - val_loss: 0.3015 - val_mae: 0.3910
Epoch 44/100
363/363 [==============================] - 2s 5ms/step - loss: 0.3021 - mae: 0.3668 - val_loss: 0.3075 - val_mae: 0.3721
Epoch 45/100
363/363 [==============================] - 2s 4ms/step - loss: 0.2807 - mae: 0.3666 - val_loss: 0.2924 - val_mae: 0.3785
Epoch 46/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2869 - mae: 0.3655 - val_loss: 0.2982 - val_mae: 0.3726
Epoch 47/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2932 - mae: 0.3652 - val_loss: 0.3101 - val_mae: 0.3876
Epoch 48/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2822 - mae: 0.3652 - val_loss: 0.2964 - val_mae: 0.3734
Epoch 49/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3265 - mae: 0.3664 - val_loss: 0.3092 - val_mae: 0.3940
Epoch 50/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2774 - mae: 0.3641 - val_loss: 0.2893 - val_mae: 0.3763
Epoch 51/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2802 - mae: 0.3642 - val_loss: 0.2853 - val_mae: 0.3649
Epoch 52/100
363/363 [==============================] - 1s 3ms/step - loss: 0.3031 - mae: 0.3649 - val_loss: 0.2959 - val_mae: 0.3842
Epoch 53/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2842 - mae: 0.3617 - val_loss: 0.2927 - val_mae: 0.3808
Epoch 54/100
363/363 [==============================] - 2s 4ms/step - loss: 0.2771 - mae: 0.3616 - val_loss: 0.3201 - val_mae: 0.3789
Epoch 55/100
363/363 [==============================] - 2s 5ms/step - loss: 0.2804 - mae: 0.3621 - val_loss: 0.2802 - val_mae: 0.3633
Epoch 56/100
363/363 [==============================] - 2s 5ms/step - loss: 0.2794 - mae: 0.3618 - val_loss: 0.2852 - val_mae: 0.3606
Epoch 57/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2826 - mae: 0.3616 - val_loss: 0.2879 - val_mae: 0.3611
Epoch 58/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2870 - mae: 0.3603 - val_loss: 0.2888 - val_mae: 0.3604
Epoch 59/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2701 - mae: 0.3588 - val_loss: 0.3046 - val_mae: 0.3935
Epoch 60/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2746 - mae: 0.3602 - val_loss: 0.2868 - val_mae: 0.3652
Epoch 61/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2863 - mae: 0.3593 - val_loss: 0.2935 - val_mae: 0.3784
Epoch 62/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2972 - mae: 0.3602 - val_loss: 0.2800 - val_mae: 0.3639
Epoch 63/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2735 - mae: 0.3597 - val_loss: 0.2776 - val_mae: 0.3603
Epoch 64/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2855 - mae: 0.3578 - val_loss: 0.2790 - val_mae: 0.3595
Epoch 65/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2804 - mae: 0.3579 - val_loss: 0.2785 - val_mae: 0.3550
Epoch 66/100
363/363 [==============================] - 2s 4ms/step - loss: 0.2819 - mae: 0.3576 - val_loss: 0.2860 - val_mae: 0.3651
Epoch 67/100
363/363 [==============================] - 2s 4ms/step - loss: 0.2670 - mae: 0.3565 - val_loss: 0.2776 - val_mae: 0.3516
Epoch 68/100
363/363 [==============================] - 2s 4ms/step - loss: 0.2974 - mae: 0.3563 - val_loss: 0.2797 - val_mae: 0.3617
Epoch 69/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2791 - mae: 0.3558 - val_loss: 0.2940 - val_mae: 0.3827
Epoch 70/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2700 - mae: 0.3559 - val_loss: 0.2774 - val_mae: 0.3522
Epoch 71/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2688 - mae: 0.3536 - val_loss: 0.2840 - val_mae: 0.3728
Epoch 72/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2675 - mae: 0.3542 - val_loss: 0.2765 - val_mae: 0.3630
Epoch 73/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2728 - mae: 0.3533 - val_loss: 0.2830 - val_mae: 0.3784
Epoch 74/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2650 - mae: 0.3536 - val_loss: 0.2768 - val_mae: 0.3632
Epoch 75/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2618 - mae: 0.3520 - val_loss: 0.2755 - val_mae: 0.3524
Epoch 76/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2658 - mae: 0.3526 - val_loss: 0.2717 - val_mae: 0.3531
Epoch 77/100
363/363 [==============================] - 2s 5ms/step - loss: 0.2639 - mae: 0.3504 - val_loss: 0.2775 - val_mae: 0.3609
Epoch 78/100
363/363 [==============================] - 2s 4ms/step - loss: 0.2598 - mae: 0.3499 - val_loss: 0.2723 - val_mae: 0.3521
Epoch 79/100
363/363 [==============================] - 2s 6ms/step - loss: 0.2608 - mae: 0.3502 - val_loss: 0.2783 - val_mae: 0.3533
Epoch 80/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2619 - mae: 0.3498 - val_loss: 0.2713 - val_mae: 0.3542
Epoch 81/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2577 - mae: 0.3488 - val_loss: 0.2769 - val_mae: 0.3565
Epoch 82/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2584 - mae: 0.3478 - val_loss: 0.2763 - val_mae: 0.3566
Epoch 83/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2569 - mae: 0.3480 - val_loss: 0.2709 - val_mae: 0.3497
Epoch 84/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2573 - mae: 0.3476 - val_loss: 0.2756 - val_mae: 0.3598
Epoch 85/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2551 - mae: 0.3470 - val_loss: 0.2725 - val_mae: 0.3563
Epoch 86/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2571 - mae: 0.3469 - val_loss: 0.2760 - val_mae: 0.3497
Epoch 87/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2557 - mae: 0.3461 - val_loss: 0.2808 - val_mae: 0.3520
Epoch 88/100
363/363 [==============================] - 2s 4ms/step - loss: 0.2565 - mae: 0.3466 - val_loss: 0.2713 - val_mae: 0.3522
Epoch 89/100
363/363 [==============================] - 2s 5ms/step - loss: 0.2543 - mae: 0.3462 - val_loss: 0.2707 - val_mae: 0.3550
Epoch 90/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2525 - mae: 0.3440 - val_loss: 0.2803 - val_mae: 0.3702
Epoch 91/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2587 - mae: 0.3457 - val_loss: 0.2728 - val_mae: 0.3568
Epoch 92/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2575 - mae: 0.3449 - val_loss: 0.3009 - val_mae: 0.3931
Epoch 93/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2555 - mae: 0.3451 - val_loss: 0.2695 - val_mae: 0.3443
Epoch 94/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2515 - mae: 0.3438 - val_loss: 0.2778 - val_mae: 0.3605
Epoch 95/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2539 - mae: 0.3441 - val_loss: 0.2857 - val_mae: 0.3533
Epoch 96/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2503 - mae: 0.3438 - val_loss: 0.2883 - val_mae: 0.3781
Epoch 97/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2494 - mae: 0.3435 - val_loss: 0.2848 - val_mae: 0.3799
Epoch 98/100
363/363 [==============================] - 1s 3ms/step - loss: 0.2508 - mae: 0.3432 - val_loss: 0.2672 - val_mae: 0.3442
Epoch 99/100
363/363 [==============================] - 1s 4ms/step - loss: 0.2505 - mae: 0.3423 - val_loss: 0.2754 - val_mae: 0.3712
Epoch 100/100
363/363 [==============================] - 2s 5ms/step - loss: 0.2508 - mae: 0.3425 - val_loss: 0.2739 - val_mae: 0.3717
162/162 [==============================] - 0s 2ms/step - loss: 0.2880 - mae: 0.3784

pd.DataFrame(ressults.history).plot(figsize=(8, 5))
plt.grid(True)
plt.gca().set_ylim(0, 1)
plt.show()

png

Keys	Action
`?`	Open this help
`n`	Next page
`p`	Previous page
`s`	Search