Generative Adversarial Network

In this notebook, we'll be building a generative adversarial network (GAN) trained on the MNIST dataset. From this, we'll be able to generate new handwritten digits!

GANs were first reported on in 2014 from Ian Goodfellow and others in Yoshua Bengio's lab. Since then, GANs have exploded in popularity. Here are a few examples to check out:

The idea behind GANs is that you have two networks, a generator $G$ and a discriminator $D$, competing against each other. The generator makes fake data to pass to the discriminator. The discriminator also sees real data and predicts if the data it's received is real or fake. The generator is trained to fool the discriminator, it wants to output data that looks as close as possible to real data. And the discriminator is trained to figure out which data is real and which is fake. What ends up happening is that the generator learns to make data that is indistiguishable from real data to the discriminator.

The general structure of a GAN is shown in the diagram above, using MNIST images as data. The latent sample is a random vector the generator uses to contruct it's fake images. As the generator learns through training, it figures out how to map these random vectors to recognizable images that can fool the discriminator.

The output of the discriminator is a sigmoid function, where 0 indicates a fake image and 1 indicates an real image. If you're interested only in generating new images, you can throw out the discriminator after training. Now, let's see how we build this thing in TensorFlow.

In [1]:
%matplotlib inline

import pickle as pkl
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt

In [2]:
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets('MNIST_data')

Successfully downloaded train-images-idx3-ubyte.gz 9912422 bytes.
Extracting MNIST_data\train-images-idx3-ubyte.gz
Successfully downloaded train-labels-idx1-ubyte.gz 28881 bytes.
Extracting MNIST_data\train-labels-idx1-ubyte.gz
Successfully downloaded t10k-images-idx3-ubyte.gz 1648877 bytes.
Extracting MNIST_data\t10k-images-idx3-ubyte.gz
Successfully downloaded t10k-labels-idx1-ubyte.gz 4542 bytes.
Extracting MNIST_data\t10k-labels-idx1-ubyte.gz

Model Inputs

First we need to create the inputs for our graph. We need two inputs, one for the discriminator and one for the generator. Here we'll call the discriminator input inputs_real and the generator input inputs_z. We'll assign them the appropriate sizes for each of the networks.

Exercise: Finish the model_inputs function below. Create the placeholders for inputs_real and inputs_z using the input sizes real_dim and z_dim respectively.

In [3]:
def model_inputs(real_dim, z_dim):
    inputs_real = tf.placeholder(tf.float32, (None, real_dim), name='inputs_real')
    inputs_z = tf.placeholder(tf.float32, (None, z_dim), name='inputs_z')
    return inputs_real, inputs_z

Generator network

Here we'll build the generator network. To make this network a universal function approximator, we'll need at least one hidden layer. We should use a leaky ReLU to allow gradients to flow backwards through the layer unimpeded. A leaky ReLU is like a normal ReLU, except that there is a small non-zero output for negative input values.

Variable Scope

Here we need to use tf.variable_scope for two reasons. Firstly, we're going to make sure all the variable names start with generator. Similarly, we'll prepend discriminator to the discriminator variables. This will help out later when we're training the separate networks.

We could just use tf.name_scope to set the names, but we also want to reuse these networks with different inputs. For the generator, we're going to train it, but also sample from it as we're training and after training. The discriminator will need to share variables between the fake and real input images. So, we can use the reuse keyword for tf.variable_scope to tell TensorFlow to reuse the variables instead of creating new ones if we build the graph again.

To use tf.variable_scope, you use a with statement:

with tf.variable_scope('scope_name', reuse=False):
    # code here

Here's more from the TensorFlow documentation to get another look at using tf.variable_scope.

Leaky ReLU

TensorFlow doesn't provide an operation for leaky ReLUs, so we'll need to make one . For this you can just take the outputs from a linear fully connected layer and pass them to tf.maximum. Typically, a parameter alpha sets the magnitude of the output for negative values. So, the output for negative input (x) values is alpha*x, and the output for positive x is x: $$ f(x) = max(\alpha * x, x) $$

Tanh Output

The generator has been found to perform the best with $tanh$ for the generator output. This means that we'll have to rescale the MNIST images to be between -1 and 1, instead of 0 and 1.

Exercise: Implement the generator network in the function below. You'll need to return the tanh output. Make sure to wrap your code in a variable scope, with 'generator' as the scope name, and pass the reuse keyword argument from the function to tf.variable_scope.

In [13]:
def generator(z, out_dim, n_units=128, reuse=False,  alpha=0.01):
    ''' Build the generator network.
        z : Input tensor for the generator
        out_dim : Shape of the generator output
        n_units : Number of units in hidden layer
        reuse : Reuse the variables with tf.variable_scope
        alpha : leak parameter for leaky ReLU
        out, logits: 
    with tf.variable_scope('generator', reuse=reuse):
        # Hidden layer
        h1 = tf.layers.dense(z, n_units, activation=None)
        # Leaky ReLU
        h1 = tf.maximum(alpha * h1, h1)
        # Logits and tanh output
        logits = tf.layers.dense(h1, out_dim, activation=None)
        out = tf.tanh(logits)
        return out


The discriminator network is almost exactly the same as the generator network, except that we're using a sigmoid output layer.

Exercise: Implement the discriminator network in the function below. Same as above, you'll need to return both the logits and the sigmoid output. Make sure to wrap your code in a variable scope, with 'discriminator' as the scope name, and pass the reuse keyword argument from the function arguments to tf.variable_scope.

In [11]:
def discriminator(x, n_units=128, reuse=False, alpha=0.01):
    ''' Build the discriminator network.
        x : Input tensor for the discriminator
        n_units: Number of units in hidden layer
        reuse : Reuse the variables with tf.variable_scope
        alpha : leak parameter for leaky ReLU
        out, logits: 
    with tf.variable_scope('discriminator', reuse=reuse):
        # Hidden layer
        h1 = tf.layers.dense(x, n_units, activation=None)
        # Leaky ReLU
        h1 = tf.maximum(alpha * h1, h1)
        logits = tf.layers.dense(h1, 1, activation=None)
        out = tf.sigmoid(logits)
        return out, logits


In [12]:
# Size of input image to discriminator
input_size = 784 # 28x28 MNIST images flattened
# Size of latent vector to generator
z_size = 100
# Sizes of hidden layers in generator and discriminator
g_hidden_size = 128
d_hidden_size = 128
# Leak factor for leaky ReLU
alpha = 0.01
# Label smoothing 
smooth = 0.1

Build network

Now we're building the network from the functions defined above.

First is to get our inputs, input_real, input_z from model_inputs using the sizes of the input and z.

Then, we'll create the generator, generator(input_z, input_size). This builds the generator with the appropriate input and output sizes.

Then the discriminators. We'll build two of them, one for real data and one for fake data. Since we want the weights to be the same for both real and fake data, we need to reuse the variables. For the fake data, we're getting it from the generator as g_model. So the real data discriminator is discriminator(input_real) while the fake discriminator is discriminator(g_model, reuse=True).

Exercise: Build the network from the functions you defined earlier.

In [14]:
# Create our input placeholders
input_real, input_z = model_inputs(input_size, z_size)

# Generator network here
g_model = generator(input_z, input_size, n_units=g_hidden_size, alpha=alpha)
# g_model is the generator output

# Disriminator network here
d_model_real, d_logits_real = discriminator(input_real, n_units=d_hidden_size, alpha=alpha)
d_model_fake, d_logits_fake = discriminator(g_model, reuse=True, n_units=d_hidden_size, alpha=alpha)

Discriminator and Generator Losses

Now we need to calculate the losses, which is a little tricky. For the discriminator, the total loss is the sum of the losses for real and fake images, d_loss = d_loss_real + d_loss_fake. The losses will by sigmoid cross-entropies, which we can get with tf.nn.sigmoid_cross_entropy_with_logits. We'll also wrap that in tf.reduce_mean to get the mean for all the images in the batch. So the losses will look something like

tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=logits, labels=labels))

For the real image logits, we'll use d_logits_real which we got from the discriminator in the cell above. For the labels, we want them to be all ones, since these are all real images. To help the discriminator generalize better, the labels are reduced a bit from 1.0 to 0.9, for example, using the parameter smooth. This is known as label smoothing, typically used with classifiers to improve performance. In TensorFlow, it looks something like labels = tf.ones_like(tensor) * (1 - smooth)

The discriminator loss for the fake data is similar. The logits are d_logits_fake, which we got from passing the generator output to the discriminator. These fake logits are used with labels of all zeros. Remember that we want the discriminator to output 1 for real images and 0 for fake images, so we need to set up the losses to reflect that.

Finally, the generator losses are using d_logits_fake, the fake image logits. But, now the labels are all ones. The generator is trying to fool the discriminator, so it wants to discriminator to output ones for fake images.

Exercise: Calculate the losses for the discriminator and the generator. There are two discriminator losses, one for real images and one for fake images. For the real image loss, use the real logits and (smoothed) labels of ones. For the fake image loss, use the fake logits with labels of all zeros. The total discriminator loss is the sum of those two losses. Finally, the generator loss again uses the fake logits from the discriminator, but this time the labels are all ones because the generator wants to fool the discriminator.

In [18]:
# Calculate losses
d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real,
                                                                                        1 - smooth)

d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake,

d_loss = tf.add(d_loss_real, d_loss_fake)

g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.ones_like(d_logits_fake)))


We want to update the generator and discriminator variables separately. So we need to get the variables for each part and build optimizers for the two parts. To get all the trainable variables, we use tf.trainable_variables(). This creates a list of all the variables we've defined in our graph.

For the generator optimizer, we only want to generator variables. Our past selves were nice and used a variable scope to start all of our generator variable names with generator. So, we just need to iterate through the list from tf.trainable_variables() and keep variables that start with generator. Each variable object has an attribute name which holds the name of the variable as a string ( == 'weights_0' for instance).

We can do something similar with the discriminator. All the variables in the discriminator start with discriminator.

Then, in the optimizer we pass the variable lists to the var_list keyword argument of the minimize method. This tells the optimizer to only update the listed variables. Something like tf.train.AdamOptimizer().minimize(loss, var_list=var_list) will only train the variables in var_list.

Exercise: Below, implement the optimizers for the generator and discriminator. First you'll need to get a list of trainable variables, then split that list into two lists, one for the generator variables and another for the discriminator variables. Finally, using AdamOptimizer, create an optimizer for each network that update the network variables separately.

In [21]:
# Optimizers
learning_rate = 0.002

# Get the trainable_variables, split into G and D parts
t_vars = tf.trainable_variables()
g_vars = [var for var in t_vars if'gen')]
d_vars = [var for var in t_vars if'dis')]

d_train_opt = tf.train.AdamOptimizer(learning_rate).minimize(d_loss, var_list=d_vars)
g_train_opt = tf.train.AdamOptimizer(learning_rate).minimize(g_loss, var_list=g_vars)


In [22]:
batch_size = 100
epochs = 100
samples = []
losses = []
saver = tf.train.Saver(var_list = g_vars)
with tf.Session() as sess:
    for e in range(epochs):
        for ii in range(mnist.train.num_examples//batch_size):
            batch = mnist.train.next_batch(batch_size)
            # Get images, reshape and rescale to pass to D
            batch_images = batch[0].reshape((batch_size, 784))
            batch_images = batch_images*2 - 1
            # Sample random noise for G
            batch_z = np.random.uniform(-1, 1, size=(batch_size, z_size))
            # Run optimizers
            _ =, feed_dict={input_real: batch_images, input_z: batch_z})
            _ =, feed_dict={input_z: batch_z})
        # At the end of each epoch, get the losses and print them out
        train_loss_d =, {input_z: batch_z, input_real: batch_images})
        train_loss_g = g_loss.eval({input_z: batch_z})
        print("Epoch {}/{}...".format(e+1, epochs),
              "Discriminator Loss: {:.4f}...".format(train_loss_d),
              "Generator Loss: {:.4f}".format(train_loss_g))    
        # Save losses to view after training
        losses.append((train_loss_d, train_loss_g))
        # Sample from generator as we're training for viewing afterwards
        sample_z = np.random.uniform(-1, 1, size=(16, z_size))
        gen_samples =
                       generator(input_z, input_size, reuse=True),
                       feed_dict={input_z: sample_z})
        samples.append(gen_samples), './checkpoints/generator.ckpt')

# Save training generator samples
with open('train_samples.pkl', 'wb') as f:
    pkl.dump(samples, f)

Epoch 1/100... Discriminator Loss: 0.3621... Generator Loss: 4.0174
Epoch 2/100... Discriminator Loss: 0.4161... Generator Loss: 4.0787
Epoch 3/100... Discriminator Loss: 1.3595... Generator Loss: 1.5809
Epoch 4/100... Discriminator Loss: 0.9823... Generator Loss: 3.4595
Epoch 5/100... Discriminator Loss: 0.8772... Generator Loss: 3.9914
Epoch 6/100... Discriminator Loss: 1.9944... Generator Loss: 1.9548
Epoch 7/100... Discriminator Loss: 0.9785... Generator Loss: 1.8205
Epoch 8/100... Discriminator Loss: 3.0750... Generator Loss: 1.9247
Epoch 9/100... Discriminator Loss: 0.9109... Generator Loss: 2.0567
Epoch 10/100... Discriminator Loss: 0.8770... Generator Loss: 2.5782
Epoch 11/100... Discriminator Loss: 0.8932... Generator Loss: 3.7445
Epoch 12/100... Discriminator Loss: 0.7054... Generator Loss: 4.0095
Epoch 13/100... Discriminator Loss: 1.3875... Generator Loss: 3.0895
Epoch 14/100... Discriminator Loss: 1.1404... Generator Loss: 2.8882
Epoch 15/100... Discriminator Loss: 1.2808... Generator Loss: 1.9609
Epoch 16/100... Discriminator Loss: 0.8562... Generator Loss: 1.4194
Epoch 17/100... Discriminator Loss: 0.8175... Generator Loss: 2.5543
Epoch 18/100... Discriminator Loss: 0.8572... Generator Loss: 2.2326
Epoch 19/100... Discriminator Loss: 0.9943... Generator Loss: 1.9610
Epoch 20/100... Discriminator Loss: 0.9836... Generator Loss: 2.2916
Epoch 21/100... Discriminator Loss: 0.6563... Generator Loss: 2.7818
Epoch 22/100... Discriminator Loss: 0.7512... Generator Loss: 2.7718
Epoch 23/100... Discriminator Loss: 0.7935... Generator Loss: 2.3430
Epoch 24/100... Discriminator Loss: 0.7192... Generator Loss: 2.9030
Epoch 25/100... Discriminator Loss: 1.0028... Generator Loss: 1.8448
Epoch 26/100... Discriminator Loss: 0.7379... Generator Loss: 3.2982
Epoch 27/100... Discriminator Loss: 0.8739... Generator Loss: 2.7935
Epoch 28/100... Discriminator Loss: 1.2539... Generator Loss: 2.1319
Epoch 29/100... Discriminator Loss: 0.9295... Generator Loss: 2.9747
Epoch 30/100... Discriminator Loss: 1.1812... Generator Loss: 1.9584
Epoch 31/100... Discriminator Loss: 0.9916... Generator Loss: 1.8704
Epoch 32/100... Discriminator Loss: 0.7283... Generator Loss: 2.5764
Epoch 33/100... Discriminator Loss: 0.9084... Generator Loss: 2.0046
Epoch 34/100... Discriminator Loss: 0.9168... Generator Loss: 1.8508
Epoch 35/100... Discriminator Loss: 1.1621... Generator Loss: 2.0742
Epoch 36/100... Discriminator Loss: 0.9815... Generator Loss: 2.7389
Epoch 37/100... Discriminator Loss: 1.0445... Generator Loss: 2.4093
Epoch 38/100... Discriminator Loss: 1.0728... Generator Loss: 1.7058
Epoch 39/100... Discriminator Loss: 0.9053... Generator Loss: 2.2478
Epoch 40/100... Discriminator Loss: 1.0950... Generator Loss: 1.7586
Epoch 41/100... Discriminator Loss: 1.3164... Generator Loss: 1.4159
Epoch 42/100... Discriminator Loss: 0.8672... Generator Loss: 2.0068
Epoch 43/100... Discriminator Loss: 0.8039... Generator Loss: 2.3359
Epoch 44/100... Discriminator Loss: 1.0841... Generator Loss: 1.8082
Epoch 45/100... Discriminator Loss: 1.1742... Generator Loss: 1.3955
Epoch 46/100... Discriminator Loss: 1.2628... Generator Loss: 1.2487
Epoch 47/100... Discriminator Loss: 1.0648... Generator Loss: 1.7636
Epoch 48/100... Discriminator Loss: 0.9785... Generator Loss: 1.8757
Epoch 49/100... Discriminator Loss: 1.0848... Generator Loss: 1.6021
Epoch 50/100... Discriminator Loss: 1.0629... Generator Loss: 1.6502
Epoch 51/100... Discriminator Loss: 1.0337... Generator Loss: 1.2975
Epoch 52/100... Discriminator Loss: 1.1327... Generator Loss: 1.8079
Epoch 53/100... Discriminator Loss: 1.0313... Generator Loss: 1.7657
Epoch 54/100... Discriminator Loss: 1.1668... Generator Loss: 1.8782
Epoch 55/100... Discriminator Loss: 1.1690... Generator Loss: 1.4636
Epoch 56/100... Discriminator Loss: 1.2680... Generator Loss: 1.6288
Epoch 57/100... Discriminator Loss: 0.9122... Generator Loss: 2.1811
Epoch 58/100... Discriminator Loss: 1.0779... Generator Loss: 1.9548
Epoch 59/100... Discriminator Loss: 1.1256... Generator Loss: 1.8592
Epoch 60/100... Discriminator Loss: 1.1600... Generator Loss: 1.6877
Epoch 61/100... Discriminator Loss: 1.1787... Generator Loss: 1.6442
Epoch 62/100... Discriminator Loss: 1.0114... Generator Loss: 1.8050
Epoch 63/100... Discriminator Loss: 1.0400... Generator Loss: 1.8546
Epoch 64/100... Discriminator Loss: 1.0071... Generator Loss: 1.8536
Epoch 65/100... Discriminator Loss: 1.0239... Generator Loss: 1.6811
Epoch 66/100... Discriminator Loss: 1.0324... Generator Loss: 1.4806
Epoch 67/100... Discriminator Loss: 1.0685... Generator Loss: 1.7235
Epoch 68/100... Discriminator Loss: 1.0138... Generator Loss: 1.7756
Epoch 69/100... Discriminator Loss: 1.2942... Generator Loss: 1.4752
Epoch 70/100... Discriminator Loss: 1.3486... Generator Loss: 1.7311
Epoch 71/100... Discriminator Loss: 1.0925... Generator Loss: 1.6634
Epoch 72/100... Discriminator Loss: 0.9613... Generator Loss: 1.9927
Epoch 73/100... Discriminator Loss: 0.9774... Generator Loss: 2.0232
Epoch 74/100... Discriminator Loss: 0.8068... Generator Loss: 2.1892
Epoch 75/100... Discriminator Loss: 1.1643... Generator Loss: 1.4880
Epoch 76/100... Discriminator Loss: 0.8750... Generator Loss: 2.4575
Epoch 77/100... Discriminator Loss: 1.0655... Generator Loss: 1.3661
Epoch 78/100... Discriminator Loss: 1.0157... Generator Loss: 1.6039
Epoch 79/100... Discriminator Loss: 1.0539... Generator Loss: 1.7016
Epoch 80/100... Discriminator Loss: 1.0578... Generator Loss: 1.4788
Epoch 81/100... Discriminator Loss: 0.9666... Generator Loss: 1.5737
Epoch 82/100... Discriminator Loss: 0.9853... Generator Loss: 2.0829
Epoch 83/100... Discriminator Loss: 0.9826... Generator Loss: 1.8983
Epoch 84/100... Discriminator Loss: 1.2212... Generator Loss: 1.5355
Epoch 85/100... Discriminator Loss: 1.0959... Generator Loss: 1.6838
Epoch 86/100... Discriminator Loss: 1.0720... Generator Loss: 1.4830
Epoch 87/100... Discriminator Loss: 1.0106... Generator Loss: 1.8722
Epoch 88/100... Discriminator Loss: 1.0186... Generator Loss: 1.6825
Epoch 89/100... Discriminator Loss: 1.2145... Generator Loss: 1.4519
Epoch 90/100... Discriminator Loss: 0.9710... Generator Loss: 1.6417
Epoch 91/100... Discriminator Loss: 1.0144... Generator Loss: 1.7431
Epoch 92/100... Discriminator Loss: 1.1103... Generator Loss: 1.8275
Epoch 93/100... Discriminator Loss: 1.0746... Generator Loss: 1.4797
Epoch 94/100... Discriminator Loss: 0.9623... Generator Loss: 1.9298
Epoch 95/100... Discriminator Loss: 0.8688... Generator Loss: 1.8336
Epoch 96/100... Discriminator Loss: 1.1272... Generator Loss: 1.4447
Epoch 97/100... Discriminator Loss: 1.0965... Generator Loss: 1.5223
Epoch 98/100... Discriminator Loss: 1.0046... Generator Loss: 1.6919
Epoch 99/100... Discriminator Loss: 1.1193... Generator Loss: 1.9348
Epoch 100/100... Discriminator Loss: 1.0225... Generator Loss: 2.0611

Training loss

Here we'll check out the training losses for the generator and discriminator.

In [23]:
%matplotlib inline

import matplotlib.pyplot as plt

In [24]:
fig, ax = plt.subplots()
losses = np.array(losses)
plt.plot(losses.T[0], label='Discriminator')
plt.plot(losses.T[1], label='Generator')
plt.title("Training Losses")

<matplotlib.legend.Legend at 0x1e618040a20>

Generator samples from training

Here we can view samples of images from the generator. First we'll look at images taken while training.

In [25]:
def view_samples(epoch, samples):
    fig, axes = plt.subplots(figsize=(7,7), nrows=4, ncols=4, sharey=True, sharex=True)
    for ax, img in zip(axes.flatten(), samples[epoch]):
        im = ax.imshow(img.reshape((28,28)), cmap='Greys_r')
    return fig, axes

In [26]:
# Load samples from generator taken while training
with open('train_samples.pkl', 'rb') as f:
    samples = pkl.load(f)

These are samples from the final training epoch. You can see the generator is able to reproduce numbers like 5, 7, 3, 0, 9. Since this is just a sample, it isn't representative of the full range of images this generator can make.

In [27]:
_ = view_samples(-1, samples)

Below I'm showing the generated images as the network was training, every 10 epochs. With bonus optical illusion!

In [28]:
rows, cols = 10, 6
fig, axes = plt.subplots(figsize=(7,12), nrows=rows, ncols=cols, sharex=True, sharey=True)

for sample, ax_row in zip(samples[::int(len(samples)/rows)], axes):
    for img, ax in zip(sample[::int(len(sample)/cols)], ax_row):
        ax.imshow(img.reshape((28,28)), cmap='Greys_r')

It starts out as all noise. Then it learns to make only the center white and the rest black. You can start to see some number like structures appear out of the noise. Looks like 1, 9, and 8 show up first. Then, it learns 5 and 3.

Sampling from the generator

We can also get completely new images from the generator by using the checkpoint we saved after training. We just need to pass in a new latent vector $z$ and we'll get new samples!

In [34]:
saver = tf.train.Saver(var_list=g_vars)
with tf.Session() as sess:
    saver.restore(sess, tf.train.latest_checkpoint('checkpoints'))
    sample_z = np.random.uniform(-1, 1, size=(16, z_size))
    gen_samples =
                   generator(input_z, input_size, reuse=True),
                   feed_dict={input_z: sample_z})
view_samples(0, [gen_samples])

INFO:tensorflow:Restoring parameters from checkpoints\generator.ckpt
(<matplotlib.figure.Figure at 0x1e9aa1b5710>,
 array([[<matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA7BB358>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AAAFB080>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA7CDBA8>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA814E80>],
        [<matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA8AAF98>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA8FDD30>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA92BB70>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA99A4E0>],
        [<matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA9A7FD0>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AAA360F0>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AAA77F28>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AAB0ECC0>],
        [<matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AAB589E8>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AABB7080>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AA4395F8>,
         <matplotlib.axes._subplots.AxesSubplot object at 0x000001E9AAC0A898>]], dtype=object))

In [ ]: