Tensorflow DNN scratch pad

dnn.py

import os
os.environ["TF_CPP_MIN_LOG_LEVEL"]="2"

import tensorflow as tf
import numpy as np

## TODO
## - Move learning rate
## - Save model, restore model
class NNModel():
    def __init__(self, layers, num_classes):
        print("In NNModel constructor")
        self.layers = layers
        self.num_classes = num_classes
        self.learning_rate = 0.1
        self.model = tf.estimator.Estimator(self.build)

    def build_layers(self, dictionary):
        tmp = dictionary["x"]
        for layer in self.layers:
            tmp = tf.layers.dense(tmp, layer)
        tmp = tf.layers.dense(tmp, self.num_classes)
        return tmp

    def train(self, input_fn, num_steps):
        self.model.train(input_fn, steps = num_steps)

    def evaluate(self, input_fn):
        return self.model.evaluate(input_fn)

    def predict(self, input_fn):
        return self.model.predict(input_fn)

    def build(self, features, labels, mode):
        logits = self.build_layers(features)

        pred_classes = tf.argmax(logits, axis=1)
        pred_probas = tf.nn.softmax(logits)

        if mode == tf.estimator.ModeKeys.PREDICT:
            return tf.estimator.EstimatorSpec(mode, predictions=pred_classes)

        loss_op = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(
            logits=logits, labels=tf.cast(labels, dtype=tf.int32)))
        optimizer = tf.train.GradientDescentOptimizer(learning_rate=self.learning_rate)
        train_op = optimizer.minimize(loss_op, global_step=tf.train.get_global_step())

        # Evaluate the accuracy of the model
        acc_op = tf.metrics.accuracy(labels=labels, predictions=pred_classes)

        # TF Estimators requires to return a EstimatorSpec, that specify
        # the different ops for training, evaluating, ...
        estim_specs = tf.estimator.EstimatorSpec(
            mode=mode,
            predictions=pred_classes,
            loss=loss_op,
            train_op=train_op,
            eval_metric_ops={'accuracy': acc_op})
        return estim_specs

test-dnn.py

import os
os.environ["TF_CPP_MIN_LOG_LEVEL"]="2"

import tensorflow as tf
import numpy as np

from dnn import NNModel

# Generates a linear relationship, then jiggle
# the original data points in order to add some
# random noise
def generate_data(num, jiggle=0.5):
    data = []
    label = []
    for i in range(num):
        r1 = np.random.randint(5)
        r2 = np.random.randint(5)

        l = (r1 + r2)
        d1 = (r1 * 1.0) + (jiggle * np.random.random())
        d2 = (r2 * 1.0) + (jiggle * np.random.random())

        if len(data) == 0:
            data = [d1, d2]
            label = [l]
        else:
            data = np.vstack((data, [d1, d2]))
            label = np.concatenate((label, [l]))
    return data, label


data, label = generate_data(40000)
print("Data", data.shape, data[0:10])
print("Label", label.shape, label[0:10])

eval_data, eval_label = generate_data(200)

model = NNModel([30, 20], 10)
batch_size = 500

# Training model
input_fn = tf.estimator.inputs.numpy_input_fn(
    x={"x": data},
    y=label,
    batch_size=batch_size,
    num_epochs=None, 
    shuffle=True)


model.train(input_fn, 1500)

# Evaluate model
eval_fn = tf.estimator.inputs.numpy_input_fn(
    x={"x": eval_data}, 
    y=eval_label,
    batch_size=batch_size, 
    shuffle=False)

eval_result = model.evaluate(eval_fn)
print(eval_result, str(eval_result))

test-mnist.py

import os
os.environ["TF_CPP_MIN_LOG_LEVEL"]="2"

import tensorflow as tf
import numpy as np

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("./data/", one_hot=False)

from dnn import NNModel


batch_size = 128
model = NNModel([20, 10], 10)

# Training model
input_fn = tf.estimator.inputs.numpy_input_fn(
    x={"x": mnist.train.images},
    y=mnist.train.labels,
    batch_size=batch_size,
    num_epochs=None, 
    shuffle=True)

model.train(input_fn, 500)

# Evaluate model
eval_fn = tf.estimator.inputs.numpy_input_fn(
    x={"x": mnist.test.images}, 
    y=mnist.test.labels,
    batch_size=batch_size, 
    shuffle=False)

eval_result = model.evaluate(eval_fn)
print(eval_result, str(eval_result))


# Prediction
randomIdx = np.random.randint(len(mnist.test.images), size=20)
sample = mnist.test.images[randomIdx,:]
sample_label = mnist.test.labels[randomIdx,]

prediction_fn = tf.estimator.inputs.numpy_input_fn(
    x={"x": sample}, 
    num_epochs=1, 
    shuffle = False)

predictions = model.predict(prediction_fn)
print("Prediction -> Actual")
for pred, label in zip(predictions, sample_label):
    print(pred, "->", label)