This repo originally started as a fork of Martin's Gorners excellent tutorial repo Tensorflow and deep learning, without a PhD, which quickly became my sandbox for experimentation with new features of tensor flow as well as new concepts from deep learning.
Since some of my files were partial copies of Martin's files, where I have added new elements while following the tutorial.
I kept the original repo in originals
directory with all the files from original repo (I just updated few imports so everyting
was working as expected).
Following files are following codelab based on the mentioned eariler tutorial:
mnist_2.0_five_layers_sigmoid.py
mnist_2.1_five_layers_relu_lrdecay.py
mnist_2.2_five_layers_relu_lrdecay_dropout.py
mnist_3.0_convolutional.py
mnist_3.1_convolutional_moar_filters.py
Remaining files are modifications of mnist_3.1_convolutional_moar_filters.py
with less and less dependencies on original
project files.
mnist_4.0_convolutional_moar_filters_batchnorm.py
- added batch normalizationmnist_4.1_convolutional_moar_filters_batchnorm_tensorboard.py
- instead of using custom visualizer, go with tensorboard for simple (scalar) metricsmnist_4.2_conv_bn_tensorboard_with_weights.py
- added plotting layer's weights, biases, activations and images to tensorboard. Based on an excellent video Hands-on TensorBoard (TensorFlow Dev Summit 2017) (+accompanying repo)mnist_5.0_keras_impl.py
- reimplementation in Kerasmnist_6.0_canned_estimator.py
- reimplementation using Estimator API (simple canned estimator, source)mnist_6.1_custom_estimator.py
- using custom estimators
halleys.py
- calculating roots of a polynomial using Halley's method (based on exercise from Intro to Tensorflow Coursera course)nyc_taxi_1.0_pandas_input.py
- nyc taxifare dataset, pandas input, canned estimatortaxi_trainer
- project with estimators, ready to use with CMLE, canned estimator, input functions, serving, etc.taxi_trainer_adv_feats
- copy of the above, but with wide & deep model, feature engineering (buckets, crosses, embeddings), updated metrics.house_value_1.0_basic.py
- predict house value using linear regression and one variable.house_value_1.0_dnn.py
- same as above, with more columns andDNNRegressor
house_value_1.0_custom.py
- same as above, with more columns and custom Estimator- [WIP]
house_value_1.0_keras.py
- using Keras model as an estimator in TF.