tony607 / keras_catvsdog_tf_estimator

Source for post "An Easy Guide to build new TensorFlow Datasets and Estimator with Keras Model"

An Easy Guide to build new TensorFlow Datasets and Estimator with Keras Model

The repo contains the jupyter notebook for 2 related the blog posts

An Easy Guide to build new TensorFlow Datasets and Estimator with Keras Model keras_estimator_vgg16-cat_vs_dog.ipynb

How to leverage TensorFlow's TFRecord to train Keras model keras_estimator_vgg16-cat_vs_dog-TFRecord.ipynb

Dependencies

Python 3.5, numpy, tensorflow, pillow

How to Run

Run the python notebook by cd into the directory in command line then run

jupyter notebook

Select any of the two *.ipynb files in the browser

Happy coding! Leave a comment in my post if you have any question.

keras_catvsdog_tf_estimator's Issues

ValueError: Shapes (?,) and (?, 1) must have the same rank

Getting ValueError: Shapes (?,) and (?, 1) must have the same rank error even when running the same code with only 2 necessary changes as mentioned in the pull request. My tensorflow version is '1.8.0'.

path issue

Hi, I like your tutorial! I just have one minor suggestion, since you are using a path that only exists on your computer, running the notebook directly gives me the error below. And the formatting of the path is not very Mac-freindly...

---

FileNotFoundError Traceback (most recent call last)
in ()
17 src = os.path.join(original_dataset_dir, fname)
18 dst = os.path.join(train_cats_dir, fname)
---> 19 shutil.copyfile(src, dst)
20 # Copy first 1000 dog images to train_dogs_dir
21 fnames = ['dog.{}.jpg'.format(i) for i in range(1000)]
/Users/Qihong/anaconda/envs/brainiak/lib/python3.6/shutil.py in copyfile(src, dst, follow_symlinks)
118 os.symlink(os.readlink(src), dst)
119 else:
--> 120 with open(src, 'rb') as fsrc:
121 with open(dst, 'wb') as fdst:
122 copyfileobj(fsrc, fdst)
FileNotFoundError: [Errno 2] No such file or directory: 'E:\Learning_Resources\Deep_learning\my_data_sets\DATASETS_IMAGE\dog_vs_cat\train/cat.0.jpg'

They are simple to solve, but it would be really nice if you can fix it!

define feature_columns to use remade estimator

Hi I've been able to adapt your model to my own data set of images. It was very informative so I thank you for putting it together. If I wanted to use one of the premade TF estimators like DNNClassifier how would I define the feature_column required for the estimator?

Extend to TF Dataset API

What about including the TF Dataset API now in core?