DCGAN on the MNIST Dataset¶

[Paper] [Notebook] [TF Implementation] [Torch Implementation]

In [1]:

import tempfile
import os
import numpy as np
import tensorflow as tf
from tensorflow.keras import layers
from matplotlib import pyplot as plt
import fastestimator as fe
from fastestimator.backend import binary_crossentropy, feed_forward
from fastestimator.dataset.data import mnist
from fastestimator.op.numpyop import LambdaOp
from fastestimator.op.numpyop.univariate import ExpandDims, Normalize
from fastestimator.op.tensorop import TensorOp
from fastestimator.op.tensorop.model import ModelOp, UpdateOp
from fastestimator.trace.io import ModelSaver

import tempfile import os import numpy as np import tensorflow as tf from tensorflow.keras import layers from matplotlib import pyplot as plt import fastestimator as fe from fastestimator.backend import binary_crossentropy, feed_forward from fastestimator.dataset.data import mnist from fastestimator.op.numpyop import LambdaOp from fastestimator.op.numpyop.univariate import ExpandDims, Normalize from fastestimator.op.tensorop import TensorOp from fastestimator.op.tensorop.model import ModelOp, UpdateOp from fastestimator.trace.io import ModelSaver

In [2]:

parameters

batch_size = 256
epochs = 50
train_steps_per_epoch = None
save_dir = tempfile.mkdtemp()
model_name = 'model_epoch_50.h5'

batch_size = 256 epochs = 50 train_steps_per_epoch = None save_dir = tempfile.mkdtemp() model_name = 'model_epoch_50.h5'

Building components

Step 1: Prepare training and define a Pipeline¶

We are loading data from tf.keras.datasets.mnist and defining a series of operations to perform on the data before the training:

In [3]:

train_data, _ = mnist.load_data()
pipeline = fe.Pipeline(
    train_data=train_data,
    batch_size=batch_size,
    ops=[
        ExpandDims(inputs="x", outputs="x"),
        Normalize(inputs="x", outputs="x", mean=1.0, std=1.0, max_pixel_value=127.5),
        LambdaOp(fn=lambda: np.random.normal(size=[100]).astype('float32'), outputs="z")
    ])

train_data, _ = mnist.load_data() pipeline = fe.Pipeline( train_data=train_data, batch_size=batch_size, ops=[ ExpandDims(inputs="x", outputs="x"), Normalize(inputs="x", outputs="x", mean=1.0, std=1.0, max_pixel_value=127.5), LambdaOp(fn=lambda: np.random.normal(size=[100]).astype('float32'), outputs="z") ])

Step 2: Create a model and FastEstimator Network¶

First, we have to define the network architecture for both our Generator and Discriminator. After defining the architecture, users are expected to feed the architecture definition, along with associated model names and optimizers, to fe.build.

In [4]:

def generator():
    model = tf.keras.Sequential()
    model.add(layers.Dense(7 * 7 * 256, use_bias=False, input_shape=(100, )))
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())
    model.add(layers.Reshape((7, 7, 256)))
    model.add(layers.Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False))
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())
    model.add(layers.Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False))
    model.add(layers.BatchNormalization())
    model.add(layers.LeakyReLU())
    model.add(layers.Conv2DTranspose(1, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh'))
    return model

def generator(): model = tf.keras.Sequential() model.add(layers.Dense(7 * 7 * 256, use_bias=False, input_shape=(100, ))) model.add(layers.BatchNormalization()) model.add(layers.LeakyReLU()) model.add(layers.Reshape((7, 7, 256))) model.add(layers.Conv2DTranspose(128, (5, 5), strides=(1, 1), padding='same', use_bias=False)) model.add(layers.BatchNormalization()) model.add(layers.LeakyReLU()) model.add(layers.Conv2DTranspose(64, (5, 5), strides=(2, 2), padding='same', use_bias=False)) model.add(layers.BatchNormalization()) model.add(layers.LeakyReLU()) model.add(layers.Conv2DTranspose(1, (5, 5), strides=(2, 2), padding='same', use_bias=False, activation='tanh')) return model

In [5]:

def discriminator():
    model = tf.keras.Sequential()
    model.add(layers.Conv2D(64, (5, 5), strides=(2, 2), padding='same', input_shape=[28, 28, 1]))
    model.add(layers.LeakyReLU())
    model.add(layers.Dropout(0.3))
    model.add(layers.Conv2D(128, (5, 5), strides=(2, 2), padding='same'))
    model.add(layers.LeakyReLU())
    model.add(layers.Dropout(0.3))
    model.add(layers.Flatten())
    model.add(layers.Dense(1))
    return model

def discriminator(): model = tf.keras.Sequential() model.add(layers.Conv2D(64, (5, 5), strides=(2, 2), padding='same', input_shape=[28, 28, 1])) model.add(layers.LeakyReLU()) model.add(layers.Dropout(0.3)) model.add(layers.Conv2D(128, (5, 5), strides=(2, 2), padding='same')) model.add(layers.LeakyReLU()) model.add(layers.Dropout(0.3)) model.add(layers.Flatten()) model.add(layers.Dense(1)) return model

In [6]:

gen_model = fe.build(model_fn=generator, optimizer_fn=lambda: tf.optimizers.Adam(1e-4))
disc_model = fe.build(model_fn=discriminator, optimizer_fn=lambda: tf.optimizers.Adam(1e-4))

gen_model = fe.build(model_fn=generator, optimizer_fn=lambda: tf.optimizers.Adam(1e-4)) disc_model = fe.build(model_fn=discriminator, optimizer_fn=lambda: tf.optimizers.Adam(1e-4))

2022-05-17 22:24:12.679107: I tensorflow/core/platform/cpu_feature_guard.cc:151] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2022-05-17 22:24:13.365903: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 32253 MB memory:  -> device: 0, name: NVIDIA A100-SXM4-40GB, pci bus id: 0000:90:00.0, compute capability: 8.0

We define the generator and discriminator losses. These can have multiple inputs and outputs.

In [7]:

class GLoss(TensorOp):
    """Compute generator loss."""
    def forward(self, data, state):
        return binary_crossentropy(y_pred=data, y_true=tf.ones_like(data), from_logits=True)

class GLoss(TensorOp): """Compute generator loss.""" def forward(self, data, state): return binary_crossentropy(y_pred=data, y_true=tf.ones_like(data), from_logits=True)

In [8]:

class DLoss(TensorOp):
    """Compute discriminator loss."""
    def forward(self, data, state):
        true_score, fake_score = data
        real_loss = binary_crossentropy(y_pred=true_score, y_true=tf.ones_like(true_score), from_logits=True)
        fake_loss = binary_crossentropy(y_pred=fake_score, y_true=tf.zeros_like(fake_score), from_logits=True)
        total_loss = real_loss + fake_loss
        return total_loss

class DLoss(TensorOp): """Compute discriminator loss.""" def forward(self, data, state): true_score, fake_score = data real_loss = binary_crossentropy(y_pred=true_score, y_true=tf.ones_like(true_score), from_logits=True) fake_loss = binary_crossentropy(y_pred=fake_score, y_true=tf.zeros_like(fake_score), from_logits=True) total_loss = real_loss + fake_loss return total_loss

fe.Network takes series of operators. Here we pass our models wrapped into ModelOps along with our loss functions and some update rules:

In [9]:

network = fe.Network(ops=[
        ModelOp(model=gen_model, inputs="z", outputs="x_fake"),
        ModelOp(model=disc_model, inputs="x_fake", outputs="fake_score"),
        GLoss(inputs="fake_score", outputs="gloss"),
        UpdateOp(model=gen_model, loss_name="gloss"),
        ModelOp(inputs="x", model=disc_model, outputs="true_score"),
        DLoss(inputs=("true_score", "fake_score"), outputs="dloss"),
        UpdateOp(model=disc_model, loss_name="dloss")
    ])

network = fe.Network(ops=[ ModelOp(model=gen_model, inputs="z", outputs="x_fake"), ModelOp(model=disc_model, inputs="x_fake", outputs="fake_score"), GLoss(inputs="fake_score", outputs="gloss"), UpdateOp(model=gen_model, loss_name="gloss"), ModelOp(inputs="x", model=disc_model, outputs="true_score"), DLoss(inputs=("true_score", "fake_score"), outputs="dloss"), UpdateOp(model=disc_model, loss_name="dloss") ])

2022-05-17 22:24:13.980469: W tensorflow/python/util/util.cc:368] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.

Step 3: Prepare Estimator and configure the training loop¶

We will define an Estimator that has four notable arguments: network, pipeline, epochs and traces. Our Network and Pipeline objects are passed here as an argument along with the number of epochs and a Trace, in this case one designed to save our model every 5 epochs.

In [10]:

traces=ModelSaver(model=gen_model, save_dir=save_dir, frequency=5)

traces=ModelSaver(model=gen_model, save_dir=save_dir, frequency=5)

In [11]:

estimator = fe.Estimator(pipeline=pipeline,
                         network=network,
                         epochs=epochs,
                         traces=traces,
                         train_steps_per_epoch=train_steps_per_epoch)

estimator = fe.Estimator(pipeline=pipeline, network=network, epochs=epochs, traces=traces, train_steps_per_epoch=train_steps_per_epoch)

Training

In [12]:

estimator.fit()

estimator.fit()

    ______           __  ______     __  _                 __            
   / ____/___ ______/ /_/ ____/____/ /_(_)___ ___  ____ _/ /_____  _____
  / /_  / __ `/ ___/ __/ __/ / ___/ __/ / __ `__ \/ __ `/ __/ __ \/ ___/
 / __/ / /_/ (__  ) /_/ /___(__  ) /_/ / / / / / / /_/ / /_/ /_/ / /    
/_/    \__,_/____/\__/_____/____/\__/_/_/ /_/ /_/\__,_/\__/\____/_/     
                                                                        

FastEstimator-Warn: the key 'y' is being pruned since it is unused outside of the Pipeline. To prevent this, you can declare the key as an input of a Trace or TensorOp.

2022-05-17 22:24:19.148391: I tensorflow/stream_executor/cuda/cuda_dnn.cc:368] Loaded cuDNN version 8100
2022-05-17 22:24:19.882754: I tensorflow/stream_executor/cuda/cuda_blas.cc:1786] TensorFloat-32 will be used for the matrix multiplication. This will only be logged once.

FastEstimator-Start: step: 1; logging_interval: 100; num_device: 1;
FastEstimator-Train: step: 1; dloss: 1.4074755; gloss: 0.6938505;
FastEstimator-Train: step: 100; dloss: 1.071872; gloss: 0.6625423; steps/sec: 96.76;
FastEstimator-Train: step: 200; dloss: 1.1390224; gloss: 0.7711286; steps/sec: 153.13;
FastEstimator-Train: step: 235; epoch: 1; epoch_time: 6.54 sec;
FastEstimator-Train: step: 300; dloss: 1.1058098; gloss: 0.8283564; steps/sec: 21.38;
FastEstimator-Train: step: 400; dloss: 1.2330828; gloss: 0.8234701; steps/sec: 137.98;
FastEstimator-Train: step: 470; epoch: 2; epoch_time: 4.88 sec;
FastEstimator-Train: step: 500; dloss: 1.0860771; gloss: 0.8769846; steps/sec: 26.53;
FastEstimator-Train: step: 600; dloss: 1.3000612; gloss: 0.76368; steps/sec: 139.42;
FastEstimator-Train: step: 700; dloss: 1.3079492; gloss: 0.7894113; steps/sec: 136.25;
FastEstimator-Train: step: 705; epoch: 3; epoch_time: 4.73 sec;
FastEstimator-Train: step: 800; dloss: 0.95769334; gloss: 1.0632424; steps/sec: 24.9;
FastEstimator-Train: step: 900; dloss: 1.0605512; gloss: 1.0276053; steps/sec: 118.47;
FastEstimator-Train: step: 940; epoch: 4; epoch_time: 5.15 sec;
FastEstimator-Train: step: 1000; dloss: 1.1515687; gloss: 1.0443377; steps/sec: 21.87;
FastEstimator-Train: step: 1100; dloss: 1.312542; gloss: 0.8080373; steps/sec: 143.04;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_5.h5
FastEstimator-Train: step: 1175; epoch: 5; epoch_time: 5.49 sec;
FastEstimator-Train: step: 1200; dloss: 1.5896232; gloss: 0.69061863; steps/sec: 24.5;
FastEstimator-Train: step: 1300; dloss: 1.0532413; gloss: 1.063014; steps/sec: 139.64;
FastEstimator-Train: step: 1400; dloss: 1.1432295; gloss: 0.9709608; steps/sec: 118.93;
FastEstimator-Train: step: 1410; epoch: 6; epoch_time: 5.16 sec;
FastEstimator-Train: step: 1500; dloss: 1.1589987; gloss: 0.88277376; steps/sec: 23.83;
FastEstimator-Train: step: 1600; dloss: 1.4223424; gloss: 0.82864374; steps/sec: 136.68;
FastEstimator-Train: step: 1645; epoch: 7; epoch_time: 5.19 sec;
FastEstimator-Train: step: 1700; dloss: 1.0954866; gloss: 1.0265985; steps/sec: 24.66;
FastEstimator-Train: step: 1800; dloss: 1.2621644; gloss: 0.8546214; steps/sec: 137.29;
FastEstimator-Train: step: 1880; epoch: 8; epoch_time: 5.04 sec;
FastEstimator-Train: step: 1900; dloss: 1.23395; gloss: 0.9362403; steps/sec: 24.59;
FastEstimator-Train: step: 2000; dloss: 1.1891618; gloss: 0.93046916; steps/sec: 141.07;
FastEstimator-Train: step: 2100; dloss: 1.1270559; gloss: 0.83571935; steps/sec: 135.63;
FastEstimator-Train: step: 2115; epoch: 9; epoch_time: 5.04 sec;
FastEstimator-Train: step: 2200; dloss: 1.0217198; gloss: 1.0431002; steps/sec: 24.59;
FastEstimator-Train: step: 2300; dloss: 1.0043646; gloss: 0.987643; steps/sec: 134.45;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_10.h5
FastEstimator-Train: step: 2350; epoch: 10; epoch_time: 5.05 sec;
FastEstimator-Train: step: 2400; dloss: 1.1894727; gloss: 0.957337; steps/sec: 26.01;
FastEstimator-Train: step: 2500; dloss: 1.5908065; gloss: 0.67834544; steps/sec: 144.04;
FastEstimator-Train: step: 2585; epoch: 11; epoch_time: 4.8 sec;
FastEstimator-Train: step: 2600; dloss: 1.1317401; gloss: 1.0065825; steps/sec: 24.88;
FastEstimator-Train: step: 2700; dloss: 1.0598496; gloss: 1.0172932; steps/sec: 131.43;
FastEstimator-Train: step: 2800; dloss: 1.2017224; gloss: 0.9210052; steps/sec: 133.42;
FastEstimator-Train: step: 2820; epoch: 12; epoch_time: 5.07 sec;
FastEstimator-Train: step: 2900; dloss: 1.1818032; gloss: 0.8824554; steps/sec: 25.56;
FastEstimator-Train: step: 3000; dloss: 1.0128953; gloss: 1.0633328; steps/sec: 123.48;
FastEstimator-Train: step: 3055; epoch: 13; epoch_time: 4.97 sec;
FastEstimator-Train: step: 3100; dloss: 1.1521642; gloss: 0.98313975; steps/sec: 24.99;
FastEstimator-Train: step: 3200; dloss: 1.2207543; gloss: 1.0411373; steps/sec: 130.05;
FastEstimator-Train: step: 3290; epoch: 14; epoch_time: 5.11 sec;
FastEstimator-Train: step: 3300; dloss: 1.0686741; gloss: 1.04258; steps/sec: 24.05;
FastEstimator-Train: step: 3400; dloss: 1.1607264; gloss: 0.9313967; steps/sec: 125.47;
FastEstimator-Train: step: 3500; dloss: 1.108815; gloss: 1.0197322; steps/sec: 128.05;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_15.h5
FastEstimator-Train: step: 3525; epoch: 15; epoch_time: 5.18 sec;
FastEstimator-Train: step: 3600; dloss: 0.84284836; gloss: 1.3197017; steps/sec: 25.89;
FastEstimator-Train: step: 3700; dloss: 1.3472433; gloss: 0.9266864; steps/sec: 131.6;
FastEstimator-Train: step: 3760; epoch: 16; epoch_time: 4.87 sec;
FastEstimator-Train: step: 3800; dloss: 1.1090488; gloss: 1.148634; steps/sec: 24.4;
FastEstimator-Train: step: 3900; dloss: 1.0149467; gloss: 1.171385; steps/sec: 134.44;
FastEstimator-Train: step: 3995; epoch: 17; epoch_time: 5.12 sec;
FastEstimator-Train: step: 4000; dloss: 1.0201095; gloss: 1.3736193; steps/sec: 24.89;
FastEstimator-Train: step: 4100; dloss: 1.1828756; gloss: 1.1176487; steps/sec: 127.13;
FastEstimator-Train: step: 4200; dloss: 0.9339646; gloss: 1.2892692; steps/sec: 129.5;
FastEstimator-Train: step: 4230; epoch: 18; epoch_time: 5.09 sec;
FastEstimator-Train: step: 4300; dloss: 1.031354; gloss: 1.1992577; steps/sec: 22.92;
FastEstimator-Train: step: 4400; dloss: 0.59360105; gloss: 1.7209172; steps/sec: 131.32;
FastEstimator-Train: step: 4465; epoch: 19; epoch_time: 5.35 sec;
FastEstimator-Train: step: 4500; dloss: 0.8413017; gloss: 1.3761427; steps/sec: 22.88;
FastEstimator-Train: step: 4600; dloss: 1.0306518; gloss: 1.3534794; steps/sec: 133.22;
FastEstimator-Train: step: 4700; dloss: 1.1800865; gloss: 1.1182059; steps/sec: 128.35;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_20.h5
FastEstimator-Train: step: 4700; epoch: 20; epoch_time: 5.41 sec;
FastEstimator-Train: step: 4800; dloss: 1.0419474; gloss: 1.090816; steps/sec: 24.75;
FastEstimator-Train: step: 4900; dloss: 0.9564374; gloss: 1.2522755; steps/sec: 136.4;
FastEstimator-Train: step: 4935; epoch: 21; epoch_time: 5.05 sec;
FastEstimator-Train: step: 5000; dloss: 1.1178182; gloss: 1.0265534; steps/sec: 26.88;
FastEstimator-Train: step: 5100; dloss: 0.9095393; gloss: 1.2775141; steps/sec: 139.1;
FastEstimator-Train: step: 5170; epoch: 22; epoch_time: 4.95 sec;
FastEstimator-Train: step: 5200; dloss: 1.1132319; gloss: 1.1454469; steps/sec: 20.47;
FastEstimator-Train: step: 5300; dloss: 1.1913971; gloss: 1.2246889; steps/sec: 73.91;
FastEstimator-Train: step: 5400; dloss: 0.94440126; gloss: 1.2502146; steps/sec: 84.12;
FastEstimator-Train: step: 5405; epoch: 23; epoch_time: 6.99 sec;
FastEstimator-Train: step: 5500; dloss: 1.0678749; gloss: 1.2804728; steps/sec: 17.85;
FastEstimator-Train: step: 5600; dloss: 1.0255647; gloss: 1.2616677; steps/sec: 80.99;
FastEstimator-Train: step: 5640; epoch: 24; epoch_time: 7.2 sec;
FastEstimator-Train: step: 5700; dloss: 0.9666496; gloss: 1.2296994; steps/sec: 19.36;
FastEstimator-Train: step: 5800; dloss: 1.116164; gloss: 0.96720326; steps/sec: 87.0;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_25.h5
FastEstimator-Train: step: 5875; epoch: 25; epoch_time: 6.74 sec;
FastEstimator-Train: step: 5900; dloss: 1.034753; gloss: 1.1733758; steps/sec: 19.73;
FastEstimator-Train: step: 6000; dloss: 1.0251997; gloss: 1.2077578; steps/sec: 80.12;
FastEstimator-Train: step: 6100; dloss: 0.9085286; gloss: 1.4258803; steps/sec: 82.07;
FastEstimator-Train: step: 6110; epoch: 26; epoch_time: 6.79 sec;
FastEstimator-Train: step: 6200; dloss: 0.91629565; gloss: 1.2635107; steps/sec: 18.95;
FastEstimator-Train: step: 6300; dloss: 0.9700071; gloss: 1.2341801; steps/sec: 81.38;
FastEstimator-Train: step: 6345; epoch: 27; epoch_time: 6.91 sec;
FastEstimator-Train: step: 6400; dloss: 1.0925424; gloss: 1.1383421; steps/sec: 19.67;
FastEstimator-Train: step: 6500; dloss: 0.886595; gloss: 1.2264949; steps/sec: 88.04;
FastEstimator-Train: step: 6580; epoch: 28; epoch_time: 6.63 sec;
FastEstimator-Train: step: 6600; dloss: 0.9399621; gloss: 1.2295549; steps/sec: 18.91;
FastEstimator-Train: step: 6700; dloss: 1.1239285; gloss: 1.0375732; steps/sec: 85.2;
FastEstimator-Train: step: 6800; dloss: 1.115287; gloss: 1.074056; steps/sec: 87.68;
FastEstimator-Train: step: 6815; epoch: 29; epoch_time: 6.77 sec;
FastEstimator-Train: step: 6900; dloss: 0.99310243; gloss: 1.0726111; steps/sec: 19.46;
FastEstimator-Train: step: 7000; dloss: 1.3016744; gloss: 1.1302423; steps/sec: 80.95;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_30.h5
FastEstimator-Train: step: 7050; epoch: 30; epoch_time: 6.59 sec;
FastEstimator-Train: step: 7100; dloss: 1.1211019; gloss: 1.0915082; steps/sec: 22.69;
FastEstimator-Train: step: 7200; dloss: 1.0433137; gloss: 1.3584124; steps/sec: 74.73;
FastEstimator-Train: step: 7285; epoch: 31; epoch_time: 6.6 sec;
FastEstimator-Train: step: 7300; dloss: 1.3104564; gloss: 1.133613; steps/sec: 18.21;
FastEstimator-Train: step: 7400; dloss: 1.195024; gloss: 1.1311966; steps/sec: 82.59;
FastEstimator-Train: step: 7500; dloss: 1.1094584; gloss: 1.1424885; steps/sec: 84.76;
FastEstimator-Train: step: 7520; epoch: 32; epoch_time: 6.99 sec;
FastEstimator-Train: step: 7600; dloss: 0.9371455; gloss: 1.1962738; steps/sec: 18.9;
FastEstimator-Train: step: 7700; dloss: 1.039571; gloss: 1.158539; steps/sec: 84.59;
FastEstimator-Train: step: 7755; epoch: 33; epoch_time: 6.9 sec;
FastEstimator-Train: step: 7800; dloss: 1.2328132; gloss: 1.0920696; steps/sec: 18.49;
FastEstimator-Train: step: 7900; dloss: 1.263283; gloss: 1.057182; steps/sec: 75.0;
FastEstimator-Train: step: 7990; epoch: 34; epoch_time: 7.16 sec;
FastEstimator-Train: step: 8000; dloss: 1.1744511; gloss: 1.0851499; steps/sec: 18.84;
FastEstimator-Train: step: 8100; dloss: 1.2414798; gloss: 0.9666091; steps/sec: 68.94;
FastEstimator-Train: step: 8200; dloss: 1.101659; gloss: 1.0953238; steps/sec: 84.85;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_35.h5
FastEstimator-Train: step: 8225; epoch: 35; epoch_time: 7.23 sec;
FastEstimator-Train: step: 8300; dloss: 1.1480453; gloss: 1.0929885; steps/sec: 18.74;
FastEstimator-Train: step: 8400; dloss: 1.0150359; gloss: 1.2104923; steps/sec: 75.67;
FastEstimator-Train: step: 8460; epoch: 36; epoch_time: 6.93 sec;
FastEstimator-Train: step: 8500; dloss: 1.2448437; gloss: 0.95777893; steps/sec: 19.26;
FastEstimator-Train: step: 8600; dloss: 1.128643; gloss: 0.9374757; steps/sec: 84.76;
FastEstimator-Train: step: 8695; epoch: 37; epoch_time: 6.76 sec;
FastEstimator-Train: step: 8700; dloss: 1.263104; gloss: 1.0629486; steps/sec: 18.54;
FastEstimator-Train: step: 8800; dloss: 1.0654749; gloss: 1.1919184; steps/sec: 81.95;
FastEstimator-Train: step: 8900; dloss: 1.0990744; gloss: 1.0925766; steps/sec: 80.36;
FastEstimator-Train: step: 8930; epoch: 38; epoch_time: 7.05 sec;
FastEstimator-Train: step: 9000; dloss: 1.3174253; gloss: 0.9213707; steps/sec: 21.57;
FastEstimator-Train: step: 9100; dloss: 1.3086524; gloss: 0.89782405; steps/sec: 87.45;
FastEstimator-Train: step: 9165; epoch: 39; epoch_time: 6.25 sec;
FastEstimator-Train: step: 9200; dloss: 1.2255262; gloss: 0.9562658; steps/sec: 18.78;
FastEstimator-Train: step: 9300; dloss: 1.2224264; gloss: 1.0891576; steps/sec: 80.81;
FastEstimator-Train: step: 9400; dloss: 1.2472556; gloss: 1.0490451; steps/sec: 76.39;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_40.h5
FastEstimator-Train: step: 9400; epoch: 40; epoch_time: 7.04 sec;
FastEstimator-Train: step: 9500; dloss: 1.2820866; gloss: 0.8918587; steps/sec: 19.16;
FastEstimator-Train: step: 9600; dloss: 1.266947; gloss: 0.9083342; steps/sec: 80.53;
FastEstimator-Train: step: 9635; epoch: 41; epoch_time: 6.89 sec;
FastEstimator-Train: step: 9700; dloss: 1.2590028; gloss: 0.91593623; steps/sec: 18.61;
FastEstimator-Train: step: 9800; dloss: 1.2626295; gloss: 0.8595179; steps/sec: 85.58;
FastEstimator-Train: step: 9870; epoch: 42; epoch_time: 6.98 sec;
FastEstimator-Train: step: 9900; dloss: 1.2414654; gloss: 0.9379455; steps/sec: 19.63;
FastEstimator-Train: step: 10000; dloss: 1.209197; gloss: 0.8486848; steps/sec: 80.11;
FastEstimator-Train: step: 10100; dloss: 1.221647; gloss: 1.0560616; steps/sec: 82.71;
FastEstimator-Train: step: 10105; epoch: 43; epoch_time: 6.79 sec;
FastEstimator-Train: step: 10200; dloss: 1.2570242; gloss: 0.9568354; steps/sec: 18.62;
FastEstimator-Train: step: 10300; dloss: 1.1466038; gloss: 1.1030976; steps/sec: 86.59;
FastEstimator-Train: step: 10340; epoch: 44; epoch_time: 6.91 sec;
FastEstimator-Train: step: 10400; dloss: 1.3359842; gloss: 0.815323; steps/sec: 18.28;
FastEstimator-Train: step: 10500; dloss: 1.436327; gloss: 0.82999104; steps/sec: 79.86;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_45.h5
FastEstimator-Train: step: 10575; epoch: 45; epoch_time: 7.27 sec;
FastEstimator-Train: step: 10600; dloss: 1.3078375; gloss: 0.9893749; steps/sec: 17.96;
FastEstimator-Train: step: 10700; dloss: 1.50738; gloss: 0.88688844; steps/sec: 79.93;
FastEstimator-Train: step: 10800; dloss: 1.2391019; gloss: 0.94254756; steps/sec: 82.19;
FastEstimator-Train: step: 10810; epoch: 46; epoch_time: 7.2 sec;
FastEstimator-Train: step: 10900; dloss: 1.2033031; gloss: 0.9017648; steps/sec: 20.95;
FastEstimator-Train: step: 11000; dloss: 1.1743715; gloss: 0.9873478; steps/sec: 80.95;
FastEstimator-Train: step: 11045; epoch: 47; epoch_time: 6.37 sec;
FastEstimator-Train: step: 11100; dloss: 1.2350852; gloss: 0.8709105; steps/sec: 19.28;
FastEstimator-Train: step: 11200; dloss: 1.3278979; gloss: 0.8681942; steps/sec: 81.55;
FastEstimator-Train: step: 11280; epoch: 48; epoch_time: 7.01 sec;
FastEstimator-Train: step: 11300; dloss: 1.2809682; gloss: 0.87823415; steps/sec: 18.17;
FastEstimator-Train: step: 11400; dloss: 1.3148129; gloss: 0.9447537; steps/sec: 80.3;
FastEstimator-Train: step: 11500; dloss: 1.2640177; gloss: 0.9070884; steps/sec: 82.19;
FastEstimator-Train: step: 11515; epoch: 49; epoch_time: 7.03 sec;
FastEstimator-Train: step: 11600; dloss: 1.209223; gloss: 0.99375534; steps/sec: 20.29;
FastEstimator-Train: step: 11700; dloss: 1.1119881; gloss: 0.991818; steps/sec: 83.43;
FastEstimator-ModelSaver: Saved model to /tmp/tmp7hg1f143/model_epoch_50.h5
FastEstimator-Train: step: 11750; epoch: 50; epoch_time: 6.52 sec;
FastEstimator-Finish: step: 11750; model1_lr: 1e-04; model_lr: 1e-04; total_time: 307.39 sec;

Inferencing

For inferencing, first we have to load the trained model weights. We will load the trained generator weights using fe.build

In [13]:

model_path = os.path.join(save_dir, model_name)
trained_model = fe.build(model_fn=generator, weights_path=model_path, optimizer_fn=lambda: tf.optimizers.Adam(1e-4))

model_path = os.path.join(save_dir, model_name) trained_model = fe.build(model_fn=generator, weights_path=model_path, optimizer_fn=lambda: tf.optimizers.Adam(1e-4))

We will the generate some images from random noise:

In [14]:

images = feed_forward(trained_model, np.random.normal(size=(16, 100)), training=False)

images = feed_forward(trained_model, np.random.normal(size=(16, 100)), training=False)

In [15]:

fe.util.BatchDisplay(image=images).show()

fe.util.BatchDisplay(image=images).show()