Initial commit (c2a480bd) · Commits · examples / Introduction to Deep Learning Models With TensorFlow

Introduction to Deep Learning Models With TensorFlow - Working Files/Working Files - README.txt

0 → 100755

+5 −0

Changes for Introduction to Deep Learning Models With TensorFlow - Working Files/Working Files - README.txt: 5 added lines, 0 removed lines.
Original line number	Diff line number	Diff line
		Due to ongoing efforts to ensure you have the best possible learning experience, we may update the working files for this course from time to time.

		We recommend that you download the latest version of these files from:

		http://www.infiniteskills.com/02492
		No newline at end of file

Introduction to Deep Learning Models With TensorFlow - Working Files/data_types_and_tf_graph.ipynb

0 → 100755

+459 −0

Changes for Introduction to Deep Learning Models With TensorFlow - Working Files/data_types_and_tf_graph.ipynb: 459 added lines, 0 removed lines.
Original line number	Diff line number	Diff line
		%% Cell type:code id: tags:

		``` python
		import tensorflow as tf
		import numpy as np
		```

		%% Cell type:code id: tags:

		``` python
		x1 = np.random.randn(100,100)
		x2 = np.random.randn(100,100)
		x3 = np.random.randn(100,100)
		```

		%% Cell type:code id: tags:

		``` python
		y = np.matmul(x1,x2)
		```

		%% Cell type:code id: tags:

		``` python
		y[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		y = np.matmul(x3,np.matmul(x2,x1))
		```

		%% Cell type:code id: tags:

		``` python
		y[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		xt1 = tf.random_normal(shape=(100,100))
		xt2 = tf.random_normal(shape=(100,100))
		xt3 = tf.random_normal(shape=(100,100))
		```

		%% Cell type:code id: tags:

		``` python
		y = tf.matmul(xt3,tf.matmul(xt2,xt1))
		```

		%% Cell type:code id: tags:

		``` python
		y
		```

		%% Cell type:code id: tags:

		``` python
		y[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		type(y)
		```

		%% Cell type:code id: tags:

		``` python
		sess = tf.Session()
		```

		%% Cell type:code id: tags:

		``` python
		out = sess.run(y)
		```

		%% Cell type:code id: tags:

		``` python
		out[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		type(out)
		```

		%% Cell type:code id: tags:

		``` python
		out1, out2 = sess.run([y,xt1])
		```

		%% Cell type:code id: tags:

		``` python
		out1[0,0], out2[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		xt4 = tf.random_normal(shape=(50,50))
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(y)[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		sess.close()
		```

		%% Cell type:code id: tags:

		``` python
		with tf.Session() as sess:
		out = sess.run(y)
		```

		%% Cell type:code id: tags:

		``` python
		out[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		sess = tf.InteractiveSession()
		```

		%% Cell type:code id: tags:

		``` python
		out = y.eval()
		```

		%% Cell type:code id: tags:

		``` python
		out[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		a = tf.Variable([0.0,1.0,2.0])
		```

		%% Cell type:code id: tags:

		``` python
		a
		```

		%% Cell type:code id: tags:

		``` python
		print(a.get_shape())
		```

		%% Cell type:code id: tags:

		``` python
		#SHOULD THROW EXCEPTION
		a.eval()
		```

		%% Cell type:code id: tags:

		``` python
		a.initializer
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(a.initializer)
		```

		%% Cell type:code id: tags:

		``` python
		a.eval()
		```

		%% Cell type:code id: tags:

		``` python
		b = tf.Variable(np.random.randn(10,10))
		```

		%% Cell type:code id: tags:

		``` python
		b
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(b.initializer)
		```

		%% Cell type:code id: tags:

		``` python
		b.eval()[0,0]
		```

		%% Cell type:code id: tags:

		``` python
		d = tf.get_variable(name="d", shape=(2,2), dtype=tf.float32)
		e = tf.get_variable(name="e", shape=(2,1), dtype=tf.float32)
		f = tf.get_variable(name="f", shape=(2,2), dtype=tf.float32)
		g = tf.get_variable(name="g", shape=(2,1), dtype=tf.float32)
		```

		%% Cell type:code id: tags:

		``` python
		h = tf.matmul(d,e) + tf.matmul(f,g)
		```

		%% Cell type:code id: tags:

		``` python
		h
		```

		%% Cell type:code id: tags:

		``` python
		init = tf.global_variables_initializer()
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(init)
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(h)
		```

		%% Cell type:code id: tags:

		``` python
		i = tf.get_variable("i", shape=(2,2), initializer=tf.constant_initializer(5.0))
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(i.initializer)
		```

		%% Cell type:code id: tags:

		``` python
		i.eval()
		```

		%% Cell type:code id: tags:

		``` python
		j = tf.get_variable("j", shape=(2,2), initializer=tf.random_uniform_initializer(maxval=1.0, minval=0.0))
		#tf.random_normal_initializer(mean=0.0, stddev=1.0)
		#tf.random_uniform_initializer(maxval=0.05, minval=-0.05)
		#Google NMT paper which uses random uniform [-0.04, 0.04] https://arxiv.org/pdf/1609.08144v2.pdf
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(j.initializer)
		```

		%% Cell type:code id: tags:

		``` python
		j.eval()
		```

		%% Cell type:code id: tags:

		``` python
		p1 = tf.placeholder(name="holder1", shape=(2,), dtype=tf.float32)
		p2 = tf.constant([5.0,3.0])
		p3 = p1 + p2
		```

		%% Cell type:code id: tags:

		``` python
		#SHOULD THROW EXCEPTION
		sess.run(p3)
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(p3, feed_dict={p1:np.array([1.0,2.0])})
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(p3, feed_dict={p1:np.array([5.0,10.0])})
		```

		%% Cell type:code id: tags:

		``` python
		placeholders = [tf.placeholder(shape=(1,), dtype=tf.float32) for i in range(5)]
		val = tf.constant(10.0)
		for i in range(5):
		val = val + placeholders[i]
		```

		%% Cell type:code id: tags:

		``` python
		feed_dict = {placeholders[i]:np.array([i]) for i in range(5)}
		```

		%% Cell type:code id: tags:

		``` python
		sess.run(val, feed_dict=feed_dict)
		```

		%% Cell type:code id: tags:

		``` python
		g = tf.get_default_graph()
		```

		%% Cell type:code id: tags:

		``` python
		g
		```

		%% Cell type:code id: tags:

		``` python
		sess.graph is g
		```

		%% Cell type:code id: tags:

		``` python
		ops = g.get_operations()
		```

		%% Cell type:code id: tags:

		``` python
		ops[:5]
		```

		%% Cell type:code id: tags:

		``` python
		[print(o.name) for o in ops[:5]]
		```

		%% Cell type:code id: tags:

		``` python
		trainable_vars = tf.trainable_variables()
		```

		%% Cell type:code id: tags:

		``` python
		[print(t.name, t.get_shape()) for t in trainable_vars]
		```

		%% Cell type:code id: tags:

		``` python
		with tf.variable_scope("model1") as scope:
		a = tf.get_variable("model", shape=(1), dtype=tf.int32, initializer=tf.constant_initializer(2))
		with tf.variable_scope("model2") as scope:
		b = tf.get_variable("model", shape=(1), dtype=tf.int32, initializer=tf.constant_initializer(3))
		c = a + b
		```

		%% Cell type:code id: tags:

		``` python
		sess.run([a.initializer, b.initializer])
		sess.run(c)
		```

		%% Cell type:code id: tags:

		``` python
		[t.name for t in tf.trainable_variables()]
		```

		%% Cell type:code id: tags:

		``` python
		with tf.variable_scope("model3", reuse=False) as scope:
		a = tf.get_variable("model", shape=(1), dtype=tf.int32, initializer=tf.constant_initializer(2))
		with tf.variable_scope("model3", reuse=True) as scope:
		b = tf.get_variable("model", shape=(1), dtype=tf.int32)
		```

		%% Cell type:code id: tags:

		``` python
		b is a
		```

		%% Cell type:code id: tags:

		``` python
		[t.name for t in tf.trainable_variables()]
		```

		%% Cell type:code id: tags:

		``` python
		with tf.variable_scope("prefix1"):
		with tf.variable_scope("prefix2"):
		tf.get_variable("model", shape=(1), dtype=tf.int32, initializer=tf.constant_initializer(2))
		```

		%% Cell type:code id: tags:

		``` python
		[t.name for t in tf.trainable_variables()]
		```

		%% Cell type:code id: tags:

		``` python
		```