new tensorflow examples

Author: MMesch

machinelearning.mm

@@ -83,9 +83,9 @@
 </hook>
 <hook NAME="AutomaticEdgeColor" COUNTER="10" RULE="ON_BRANCH_CREATION"/>
 <hook NAME="accessories/plugins/AutomaticLayout.properties" VALUE="ALL"/>
-<node TEXT="model" POSITION="right" ID="ID_1486859491" CREATED="1489148242353" MODIFIED="1489149116450" HGAP_QUANTITY="14.749999977648258 pt" VSHIFT_QUANTITY="-58.49999825656419 pt">
+<node TEXT="model" POSITION="right" ID="ID_1486859491" CREATED="1489148242353" MODIFIED="1490358639221" HGAP_QUANTITY="14.749999977648258 pt" VSHIFT_QUANTITY="-58.49999825656419 pt">
 <edge COLOR="#00ffff"/>
-<node TEXT="regression (fit Y = M(X))" ID="ID_848707408" CREATED="1489148691975" MODIFIED="1489150550042">
+<node TEXT="regression (fit Y = M(X))" ID="ID_848707408" CREATED="1489148691975" MODIFIED="1490358639217" HGAP_QUANTITY="14.749999977648258 pt" VSHIFT_QUANTITY="-26.99999919533732 pt">
 <node TEXT="nearest neighbor" ID="ID_192120717" CREATED="1489150509095" MODIFIED="1489150523538"/>
 <node TEXT="decision tree" ID="ID_1981317805" CREATED="1489150524625" MODIFIED="1489150527641"/>
 <node TEXT="Lasso regression" ID="ID_503979546" CREATED="1489150558163" MODIFIED="1489150564677"/>

tensorflow_simple/linear_regression.py

@@ -0,0 +1,60 @@
+#!/usr/bin/env python
+# -*- coding: utf8 -*-
+import numpy as np
+import matplotlib.pyplot as plt
+import tensorflow as tf
+
+
+def main():
+    # ---- dataset ----
+    npoints = 400
+    x0 = np.linspace(0., 10 * np.pi, npoints)
+    noise = np.random.normal(scale=0.1, size=npoints)
+    y0 = np.sin(x0) + noise
+    # y0 = noise
+
+    # ---- definitions ----
+    # Model
+    x = tf.placeholder(tf.float32, [None], name='x')
+    y_label = tf.placeholder(tf.float32, [None], name='y')
+
+    m = tf.Variable(tf.ones([1]))
+    b = tf.Variable(tf.ones([1]))
+
+    y = m * x + b
+
+    # cost function
+    cost = tf.reduce_mean(tf.square(y_label - y))
+
+    # optimizer
+    stepsize = 0.001
+    optimizer = tf.train.GradientDescentOptimizer(stepsize).minimize(cost)
+
+    # ---- run tensorflow ----
+    init = tf.initialize_all_variables()
+
+    with tf.Session() as sess:
+        sess.run(init)
+
+        for isample in range(10000):
+            optimizer.run(feed_dict={x: x0, y_label: y0})
+
+        y_predicted = sess.run(y, feed_dict={x: x0})
+        m_optimized = sess.run(m)
+        b_optimized = sess.run(b)
+
+        # write graph file
+        writer = tf.train.SummaryWriter('./', sess.graph)
+        writer.close()
+
+    print(m_optimized, b_optimized)
+    # ---- plotting ----
+    fig, axes = plt.subplots()
+    axes.plot(x0, y0)
+    axes.plot(x0, y_predicted)
+
+    plt.show()
+
+
+if __name__ == "__main__":
+    main()

tensorflow_simple/neuralnetwork_regression.py

@@ -0,0 +1,65 @@
+#!/usr/bin/env python
+# -*- coding: utf8 -*-
+import numpy as np
+import matplotlib.pyplot as plt
+import tensorflow as tf
+
+
+def main():
+    # ---- dataset ----
+    npoints = 400
+    x0 = np.linspace(0., 10 * np.pi, npoints)
+    noise = np.random.normal(scale=0.1, size=npoints)
+    y0 = (np.sin(x0) + noise)
+    # y0 = noise
+
+    # ---- definitions ----
+    # Model
+    nnetwork = 400
+
+    x = tf.placeholder(tf.float32, [None, 1], name='x')
+    y_label = tf.placeholder(tf.float32, [None, 1], name='y')
+
+    W1 = tf.Variable(tf.random_uniform([1, nnetwork], -1.0, 1.0))
+    b1 = tf.Variable(tf.zeros([nnetwork]))
+
+    h1 = tf.nn.tanh(tf.matmul(x, W1) + b1)
+
+
+    W2 = tf.Variable(tf.random_uniform([nnetwork, 1], -nnetwork, nnetwork))
+    b2 = tf.Variable(tf.zeros([1]))
+
+    y = tf.matmul(h1, W2) + b2
+
+    # cost function
+    cost = tf.reduce_mean(tf.square(y_label - y))
+
+    # optimizer
+    stepsize = 0.001
+    optimizer = tf.train.GradientDescentOptimizer(stepsize).minimize(cost)
+
+    # ---- run tensorflow ----
+    init = tf.initialize_all_variables()
+
+    with tf.Session() as sess:
+        sess.run(init)
+
+        for isample in range(10000):
+            optimizer.run(feed_dict={x: x0[:, None], y_label: y0[:, None]})
+
+        y_predicted = sess.run(y, feed_dict={x: x0[:, None]})
+
+        # write graph file
+        writer = tf.train.SummaryWriter('./', sess.graph)
+        writer.close()
+
+    # ---- plotting ----
+    fig, axes = plt.subplots()
+    axes.plot(x0, y0, 'o')
+    axes.plot(x0, y_predicted)
+
+    plt.show()
+
+
+if __name__ == "__main__":
+    main()