Switch regression examples to use units of 1000$

PiperOrigin-RevId: 169411781

4 files changed, 32 insertions, 5 deletions

diff --git a/tensorflow/examples/get_started/regression/dnn_regression.py b/tensorflow/examples/get_started/regression/dnn_regression.py
index 7aa3659139..951c93b52e 100644
--- a/tensorflow/examples/get_started/regression/dnn_regression.py
+++ b/tensorflow/examples/get_started/regression/dnn_regression.py
@@ -23,6 +23,7 @@ import tensorflow as tf
 import imports85  # pylint: disable=g-bad-import-order
 
 STEPS = 5000
+PRICE_NORM_FACTOR = 1000
 
 
 def main(argv):
@@ -30,6 +31,13 @@ def main(argv):
   assert len(argv) == 1
   (train, test) = imports85.dataset()
 
+  # Switch the labels to units of thousands for better convergence.
+  def normalize_price(features, labels):
+    return features, labels / PRICE_NORM_FACTOR
+
+  train = train.map(normalize_price)
+  test = test.map(normalize_price)
+
   # Build the training input_fn.
   def input_train():
     return (
@@ -86,7 +94,8 @@ def main(argv):
 
   # Convert MSE to Root Mean Square Error (RMSE).
   print("\n" + 80 * "*")
-  print("\nRMS error for the test set: ${:.0f}".format(average_loss**0.5))
+  print("\nRMS error for the test set: ${:.0f}"
+        .format(PRICE_NORM_FACTOR * average_loss**0.5))
 
   print()
 
diff --git a/tensorflow/examples/get_started/regression/imports85.py b/tensorflow/examples/get_started/regression/imports85.py
index 41e77222ce..c165f0175d 100644
--- a/tensorflow/examples/get_started/regression/imports85.py
+++ b/tensorflow/examples/get_started/regression/imports85.py
@@ -91,7 +91,7 @@ def dataset(y_name="price", train_fraction=0.7):
     """Convert a csv line into a (features_dict,label) pair."""
     # Decode the line to a tuple of items based on the types of
     # csv_header.values().
-    items = tf.decode_csv(line, defaults.values())
+    items = tf.decode_csv(line, list(defaults.values()))
 
     # Convert the keys and items to a dict.
     pairs = zip(defaults.keys(), items)
diff --git a/tensorflow/examples/get_started/regression/linear_regression.py b/tensorflow/examples/get_started/regression/linear_regression.py
index dd44077663..74651e7446 100644
--- a/tensorflow/examples/get_started/regression/linear_regression.py
+++ b/tensorflow/examples/get_started/regression/linear_regression.py
@@ -24,6 +24,7 @@ import tensorflow as tf
 import imports85  # pylint: disable=g-bad-import-order
 
 STEPS = 1000
+PRICE_NORM_FACTOR = 1000
 
 
 def main(argv):
@@ -31,6 +32,13 @@ def main(argv):
   assert len(argv) == 1
   (train, test) = imports85.dataset()
 
+  # Switch the labels to units of thousands for better convergence.
+  def to_thousands(features, labels):
+    return features, labels / PRICE_NORM_FACTOR
+
+  train = train.map(to_thousands)
+  test = test.map(to_thousands)
+
   # Build the training input_fn.
   def input_train():
     return (
@@ -67,7 +75,8 @@ def main(argv):
 
   # Convert MSE to Root Mean Square Error (RMSE).
   print("\n" + 80 * "*")
-  print("\nRMS error for the test set: ${:.0f}".format(average_loss**0.5))
+  print("\nRMS error for the test set: ${:.0f}"
+        .format(PRICE_NORM_FACTOR * average_loss**0.5))
 
   # Run the model in prediction mode.
   input_dict = {
@@ -85,7 +94,7 @@ def main(argv):
            "Highway: {: 0d}mpg, "
            "Prediction: ${: 9.2f}")
     msg = msg.format(input_dict["curb-weight"][i], input_dict["highway-mpg"][i],
-                     prediction["predictions"][0])
+                     PRICE_NORM_FACTOR * prediction["predictions"][0])
 
     print("    " + msg)
   print()
diff --git a/tensorflow/examples/get_started/regression/linear_regression_categorical.py b/tensorflow/examples/get_started/regression/linear_regression_categorical.py
index 38ecfada9d..860d0e437c 100644
--- a/tensorflow/examples/get_started/regression/linear_regression_categorical.py
+++ b/tensorflow/examples/get_started/regression/linear_regression_categorical.py
@@ -23,6 +23,7 @@ import tensorflow as tf
 import imports85  # pylint: disable=g-bad-import-order
 
 STEPS = 1000
+PRICE_NORM_FACTOR = 1000
 
 
 def main(argv):
@@ -30,6 +31,13 @@ def main(argv):
   assert len(argv) == 1
   (train, test) = imports85.dataset()
 
+  # Switch the labels to units of thousands for better convergence.
+  def normalize_price(features, labels):
+    return features, labels / PRICE_NORM_FACTOR
+
+  train = train.map(normalize_price)
+  test = test.map(normalize_price)
+
   # Build the training input_fn.
   def input_train():
     return (
@@ -91,7 +99,8 @@ def main(argv):
 
   # Convert MSE to Root Mean Square Error (RMSE).
   print("\n" + 80 * "*")
-  print("\nRMS error for the test set: ${:.0f}".format(average_loss**0.5))
+  print("\nRMS error for the test set: ${:.0f}"
+        .format(PRICE_NORM_FACTOR * average_loss**0.5))
 
   print()