Skip to content

Commit

Permalink
notebook cleanup (#75)
Browse files Browse the repository at this point in the history 
Notebook cleanup.
  • Loading branch information
@amygdala
amygdala committed Sep 28, 2017
1 parent 49d5cce commit c62cfa3
Showing 1 changed file with 72 additions and 101 deletions.
173 changes: 72 additions & 101 deletions workshop_sections/mnist_series/mnist_estimator.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,48 +30,64 @@
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"metadata": {},
"outputs": [],
"source": [
"import tensorflow as tf\n",
"from tensorflow.examples.tutorials.mnist import input_data\n",
"import numpy as np\n",
"import os\n",
"import time\n",
"\n",
"print(tf.__version__)\n",
"\n",
"print(tf.__version__)"
"# define a utility function for generating a new directory in which to save \n",
"# model information, so multiple training runs don't stomp on each other.\n",
"def get_new_path(name=\"\"):\n",
" base=\"/tmp/tfmodels/mnist_estimators\"\n",
" logpath = os.path.join(base, name + \"_\" + str(int(time.time())))\n",
" print(\"Logging to {}\".format(logpath))\n",
" return logpath"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true,
"scrolled": true
},
"outputs": [],
"source": [
"# Train and evaluate\n",
"# Train\n",
"\n",
"DATA_SETS = input_data.read_data_sets(\n",
" \"/tmp/fashion-mnist\")\n",
" \"/tmp/MNIST_data\")\n",
"\n",
"feature_columns = [tf.feature_column.numeric_column(\n",
" \"pixels\", shape=784)]\n",
"\n",
"m = tf.estimator.LinearClassifier(\n",
" feature_columns=feature_columns, \n",
" n_classes=10,\n",
" model_dir=\"/tmp/tfmodels/mnist_estimators/fashion_linear\")\n",
" model_dir=get_new_path(\"linear\")\n",
" )\n",
"\n",
"train_input_fn = tf.estimator.inputs.numpy_input_fn(\n",
" x={'pixels': DATA_SETS.train.images},\n",
" y=DATA_SETS.train.labels.astype(np.int64),\n",
" batch_size=100,\n",
" num_epochs=3,\n",
" shuffle=True)\n",
"m.train(input_fn=train_input_fn)\n",
"\n",
"m.train(input_fn=train_input_fn)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"# Evaluate\n",
"eval_input_fn = tf.estimator.inputs.numpy_input_fn(\n",
" x={'pixels': DATA_SETS.test.images},\n",
" y=DATA_SETS.test.labels.astype(np.int64),\n",
Expand All @@ -80,14 +96,13 @@
" shuffle=False)\n",
"results = m.evaluate(input_fn=eval_input_fn)\n",
"\n",
"print(results)\n"
"print(results)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true,
"scrolled": true
},
"outputs": [],
Expand All @@ -112,7 +127,6 @@
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true,
"scrolled": true
},
"outputs": [],
Expand Down Expand Up @@ -158,7 +172,7 @@
"from tensorflow.examples.tutorials.mnist import input_data\n",
"\n",
"# comment out for less info during the training runs.\n",
"tf.logging.set_verbosity(tf.logging.WARN)\n",
"tf.logging.set_verbosity(tf.logging.INFO)\n",
"\n",
"# confirm what version of TensorFlow you are running\n",
"print('Running TensorFlow version {}'.format(tf.__version__))"
Expand All @@ -172,20 +186,11 @@
"source": [
"# Set locations of data files\n",
"MNIST_DATA_DIR = \"/tmp/MNIST_data\"\n",
"FASHION_DATA_DIR = \"/tmp/fashion-mnist\"\n",
"# Select your choice of dataset\n",
"DATA_DIR = FASHION_DATA_DIR\n",
"# edit the following to reflect where you put the Fashion-MNIST local dir\n",
"FASHION_DATA_DIR = \"your-fashion-mnist-data\" \n",
"\n",
"# read in data, downloading first as necessary\n",
"DATA_SETS = input_data.read_data_sets(DATA_DIR)\n",
"\n",
"# define a utility function for generating a new directory in which to save \n",
"# model information, so multiple training runs don't stomp on each other.\n",
"def get_new_path(name=\"\"):\n",
" base=\"/tmp/tfmodels/mnist_estimators\"\n",
" logpath = os.path.join(base, name + \"_\" + str(int(time.time())))\n",
" print(\"Logging to {}\".format(logpath))\n",
" return logpath"
"DATA_SETS = input_data.read_data_sets(MNIST_DATA_DIR)"
]
},
{
Expand Down Expand Up @@ -227,7 +232,9 @@
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"def define_and_run_linear_classifier(num_steps, \n",
Expand All @@ -247,7 +254,9 @@
" DATA_SETS.train, \n",
" epochs=3, \n",
" shuffle=True,\n",
" batch_size=batch_size))\n",
" batch_size=batch_size),\n",
" steps=num_steps\n",
" )\n",
" \n",
" print(\"Finished training.\")\n",
" \n",
Expand Down Expand Up @@ -302,14 +311,16 @@
" DATA_SETS.train, \n",
" epochs=3, \n",
" shuffle=True,\n",
" batch_size=batch_size))\n",
" batch_size=batch_size),\n",
" steps=num_steps)\n",
"\n",
" print(\"Finished running the deep training via the train() method\")\n",
" \n",
" accuracy_score = classifier.evaluate(input_fn=generate_input_fn(\n",
" DATA_SETS.test, batch_size=batch_size, shuffle=False, epochs=1))['accuracy']\n",
"\n",
" print('DNN Classifier Accuracy: {0:f}'.format(accuracy_score))"
" print('DNN Classifier Accuracy: {0:f}'.format(accuracy_score))\n",
" return classifier"
]
},
{
Expand All @@ -336,6 +347,13 @@
"# With 1000 steps and a batch size of 40, we see accuracy of approx 91% for MNIST"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"You can point your model to an existing checkpoint directory rather than having it start from scratch:"
]
},
{
"cell_type": "code",
"execution_count": null,
Expand All @@ -344,9 +362,9 @@
},
"outputs": [],
"source": [
"model_path = \"/tmp/tfmodels/mnist_estimators/linear_1505895781\" # This is an example\n",
"for i in range(0,5):\n",
" define_and_run_linear_classifier(500, model_path, batch_size=40)"
"# model_path = \"/tmp/tfmodels/mnist_estimators/linear_1505895781\" # This is an example\n",
"# for i in range(0,5):\n",
"# define_and_run_linear_classifier(500, model_path, batch_size=40)"
]
},
{
Expand All @@ -365,8 +383,8 @@
"outputs": [],
"source": [
"print(\"Running DNN classifier with .1 learning rate...\")\n",
"define_and_run_dnn_classifier(\n",
" num_steps=500, \n",
"classifier = define_and_run_dnn_classifier(\n",
" num_steps=2000, \n",
" logdir=get_new_path(\"deep01\"), \n",
" lr=.1)\n",
"# With 2000 steps and a batch size of 40, we see accuracy of approx 95% on MNIST"
Expand All @@ -376,28 +394,21 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"Let's loop through the training-eval loop a couple of times, so we get more accuracy readings. Make a for-loop and provide a stable path for your model, which will allow continuous training-eval loops."
"If you downloaded Fashion-MNIST, let's see what MNIST and Fashion-MNIST results look like side by side. Change the path of the DATA_DIR to point to your fashion-mnist dataset, and run the training again. Be sure to change your model path."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true,
"scrolled": true
},
"outputs": [],
"source": [
"model_path = \"/tmp/tfmodels/mnist_estimators/deep01_1505896039\" # This is an example\n",
"for i in range(0,5):\n",
" define_and_run_dnn_classifier(500, model_path, lr=.01)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"outputs": [],
"source": [
"Let's see what MNIST and fashion-mnist look like side by side. Change the path of the DATA_DIR to point to your fashion-mnist dataset, and run the training again. Be sure to change your model path."
"DATA_SETS = input_data.read_data_sets(FASHION_DATA_DIR)\n",
"print(\"Running DNN classifier with Fashion-MNIST data and a .1 learning rate...\")\n",
"fclassifier = define_and_run_dnn_classifier(\n",
" num_steps=2000, \n",
" logdir=get_new_path(\"deep01f\"), \n",
" lr=.1)"
]
},
{
Expand All @@ -411,16 +422,14 @@
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true,
"scrolled": true
},
"outputs": [],
"source": [
"print(\"Running DNN classifier with .5 learning rate...\")\n",
"classifier = define_and_run_dnn_classifier(2000, \n",
" get_new_path(\"deep05\"), \n",
" lr=.5)\n",
"# With 2000 steps and a batch size of 40, we see accuracy of approx 91%, though sometimes it does not converge at all."
"classifier5 = define_and_run_dnn_classifier(2000, \n",
" get_new_path(\"deep05\"), \n",
" lr=.5)\n"
]
},
{
Expand All @@ -432,6 +441,8 @@
"```sh\n",
"$ tensorboard --logdir=/tmp/tfmodels/mnist_estimators\n",
"```\n",
"Look for it at localhost:6006\n",
"\n",
"Or run the following (select Kernel --> Interrupt from the menu when you're done):"
]
},
Expand All @@ -447,24 +458,10 @@
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"cell_type": "markdown",
"metadata": {},
"source": [
"def generate_input_fn(dataset, batch_size=BATCH_SIZE, \n",
" shuffle=False, epochs=None):\n",
" X = dataset.images\n",
" Y = dataset.labels.astype(numpy.int64)\n",
" return tf.estimator.inputs.numpy_input_fn(\n",
" x={'pixels': X},\n",
" y=Y,\n",
" batch_size=batch_size,\n",
" num_epochs=epochs,\n",
" shuffle=shuffle\n",
" )"
"We can again make some predictions using our trained models. Assuming you ran the Fashion-MNIST training, `DATA_SETS` now points to that dataset."
]
},
{
Expand All @@ -482,25 +479,16 @@
"metadata": {},
"outputs": [],
"source": [
"model_path = \"/tmp/tfmodels/mnist_estimators/deep01_1506395529\" # This is an example\n",
"\n",
"feature_columns = [tf.feature_column.numeric_column(\n",
" \"pixels\", shape=784)]\n",
"classifier = tf.estimator.DNNClassifier(\n",
" feature_columns=feature_columns, \n",
" n_classes=10,\n",
" hidden_units=[200, 100, 50],\n",
" optimizer=tf.train.ProximalAdagradOptimizer(learning_rate=0.1),\n",
" model_dir=model_path)\n",
" \n",
"X = DATA_SETS.test.images[5000:5005]\n",
"predict_input_fn = tf.estimator.inputs.numpy_input_fn( \n",
" x={'pixels': X},\n",
" batch_size=1,\n",
" num_epochs=1,\n",
" shuffle=False)\n",
"\n",
"predictions = classifier.predict(input_fn=predict_input_fn)\n",
"# if you did not run the Fashion-MNIST training, edit the following to point\n",
"# to 'classifier' instead (the 'regular' MNIST model).\n",
"predictions = fclassifier.predict(input_fn=predict_input_fn)\n",
"\n",
"for prediction in predictions:\n",
" print(\"Predictions: {} with probabilities {}\\n\".format(prediction[\"classes\"], prediction[\"probabilities\"]))"
Expand Down Expand Up @@ -544,23 +532,6 @@
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 2",
"language": "python",
"name": "python2"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 2
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython2",
"version": "2.7.13"
}
},
"nbformat": 4,
"nbformat_minor": 1
Expand Down

0 comments on commit c62cfa3

Please sign in to comment.