Using tf.contrib.learn to solve basic logistic classifier

Question

I am learning about tf.contrib.learn in Tensorflow, and am using a self-made exercise. The exercise is to classify three regions as follows, with x1 and x2 as inputs, and the labels are triangles/circles/crosses:

My code is able to fit the data, and evaluate it. However, I cannot seem to get predictions to work. Code is below. Any ideas?

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import argparse
import sys
import tempfile

from six.moves import urllib

import pandas as pd
import tensorflow as tf
import numpy as np

FLAGS = None

myImportedDatax1_np = np.array([[.1],[.1],[.2],[.2],[.4],[.4],[.5],[.5],[.1],[.1],[.2],[.2]],dtype=float)
myImportedDatax2_np = np.array([[.1],[.2],[.1],[.2],[.1],[.2],[.1],[.2],[.4],[.5],[.4],[.5]],dtype=float)
combined_Imported_Data_x = np.append(myImportedDatax1_np, myImportedDatax2_np, axis=1)
myImportedDatay_np = np.array([[0],[0],[0],[0],[1],[1],[1],[1],[2],[2],[2],[2]],dtype=int)

def build_estimator(model_dir, model_type):
  x1 = tf.contrib.layers.real_valued_column("x1")
  x2 = tf.contrib.layers.real_valued_column("x2")

  wide_columns = [x1, x2]
  m = tf.contrib.learn.LinearClassifier(model_dir=model_dir, feature_columns=wide_columns)
  return m

def input_fn(input_batch, output_batch):
  inputs = {"x1": tf.constant(input_batch[:,0]), "x2": tf.constant(input_batch[:,1])}
  label = tf.constant(output_batch)
  print(inputs)
  print(label)
  print(input_batch)
  # Returns the feature columns and the label.
  return inputs, label

def train_and_eval(model_dir, model_type, train_steps, train_data, test_data):
  model_dir = tempfile.mkdtemp() if not model_dir else model_dir
  print("model directory = %s" % model_dir)
  m = build_estimator(model_dir, model_type)
  m.fit(input_fn=lambda: input_fn(combined_Imported_Data_x, myImportedDatay_np), steps=train_steps)
  results = m.evaluate(input_fn=lambda: input_fn(np.array([[.4, .1],[.4, .2]], dtype=float), np.array([[0], [0]], dtype=int)), steps=1)
  for key in sorted(results):
    print("%s: %s" % (key, results[key]))
  predictions = list(m.predict(input_fn=({"x1": tf.constant([[.1]]),"x2": tf.constant([[.1]])})))
 # print(predictions)

def main(_):
  train_and_eval(FLAGS.model_dir, FLAGS.model_type, FLAGS.train_steps,
                 FLAGS.train_data, FLAGS.test_data)

if __name__ == "__main__":
  parser = argparse.ArgumentParser()
  parser.register("type", "bool", lambda v: v.lower() == "true")
  parser.add_argument(
      "--model_dir",
      type=str,
      default="",
      help="Base directory for output models."
  )
  parser.add_argument(
      "--model_type",
      type=str,
      default="wide_n_deep",
      help="Valid model types: {'wide', 'deep', 'wide_n_deep'}."
  )
  parser.add_argument(
      "--train_steps",
      type=int,
      default=200,
      help="Number of training steps."
  )
  parser.add_argument(
      "--train_data",
      type=str,
      default="",
      help="Path to the training data."
  )
  parser.add_argument(
      "--test_data",
      type=str,
      default="",
      help="Path to the test data."
  )
  FLAGS, unparsed = parser.parse_known_args()
  tf.app.run(main=main, argv=[sys.argv[0]] + unparsed)

amo-ej1 · Accepted Answer · 2017-07-26 05:56:46Z

1

To fix this concrete issue you can add the following input function which is similar to the existing one, except that it returns None as a second element in the tuple

def input_fn_predict():
  inputs = {"x1": tf.constant([0.1]), "x2": tf.constant([0.2])}
  print(inputs)
  return inputs, None

In a next phase you can invoke it with:

predictions = list(m.predict(input_fn=lambda: input_fn_predict()))

And if you comment out your print, then this should work.

edited Jul 26, 2017 at 5:56

answered Jul 24, 2017 at 3:37

amo-ej1

3,30728 silver badges35 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

Rowan Gontier Over a year ago

I tried:

def input_fn_predict():  inputs = {"x1": tf.constant([[0.2], [0.4], [.5]]), "x2": tf.constant([[0.1], [0.1], [.1]])}   return inputs

with code: predictions = list(m.predict(input_fn=lambda: input_fn_predict())) It runs without errors. However, the predictions are all "1", which is wrong. I think the model has fitted okay, because the evaluate function behaves as expected.

amo-ej1 Over a year ago

I updated my answer as there was a copy paste error. Regarding the correctness, when you create the LinearClassifier you do not pass the n_classes parameter, indicating you want a binary classifier, which will not work as you your problem set consists out of three classes by nature.

Rowan Gontier Over a year ago

Works fine now.

Collectives™ on Stack Overflow

Using tf.contrib.learn to solve basic logistic classifier

1 Answer 1

3 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

3 Comments

Your Answer

Sign up or log in

Post as a guest

Related