Merge pull request lukas#17 from lukas/feature/wandb

Move things around

.gitignore

@@ -0,0 +1 @@
+wandb-debug.log

README.md

@@ -33,38 +33,37 @@ If you've never used pandas or numpy - they are great tools and I use them heavi
 ## Installation
 Before the class, please run the following commands to install the prerequisite code.
 
-### Linux and Mac OS X
-#### Install python
+### Windows
 
-You can download python from https://www.python.org/downloads/.  There are more detailed instructions for windows installation at https://www.howtogeek.com/197947/how-to-install-python-on-windows/.  
+#### Git
 
-The material should work with python 2 or 3.  *On Windows, you need to install thre 64 bit version of python 3.5 or 3.6 in order to install tensorflow*.
+Install git: https://git-scm.com/download/win
+
+#### Anaconda
+
+Install [anaconda](https://repo.continuum.io/archive/Anaconda3-4.4.0-Windows-x86_64.exe)
+
+Try running the following from the command prompt:
 
-#### Clone this github repository
 ```
-git clone https://github.com/lukas/ml-class.git
-cd ml-class
+python --version
 ```
 
-If you get an error message here, most likely you don't have git installed.  Go to https://www.atlassian.com/git/tutorials/install-git for intructions on installing git.
+You should see something like
 
-#### Install necessary pip libraries
 ```
-pip install pandas
-pip install scikit-learn
-pip install tensorflow
-pip install keras
+Python 3.6.1 :: Anaconda 4.4.0 (64-bit)
 ```
 
-#### Install python libraries for optional material
+If don't see "Anaconda" in the output, search for "anaconda prompt" from the start menu and enter your command prompt this way.   It's also best to use a virtual environment to keep your packages silo'ed.  Do so with:
+
 ```
-pip install h5py
-pip install flask
-pip install scikit-image
-pip install scipy
-pip install pillow
+conda create -n ml-class python=3.6
+activate ml-class
 ```
 
+Whenever you start a new terminal, you will need to call `activate ml-class`.
+
 #### Common problems
 
 The most common problem is an old version of python.  Its easy to have multiple versions of python installed at once and Macs in particular come with a default version of python that is too old to install tensorflow.
@@ -77,42 +76,53 @@ python --version
 
 If your version is less than 2.7.12, you have a version issue.  Try reinstalling python 2.
 
-### Windows
-
-#### Git
-
-Install git from https://git-for-windows.github.io/
-
-#### Anaconda
-
-Install anaconda from https://www.anaconda.com/download/#download
-
-Add Anaconda3 and Anaconda3/Scripts to your path environmental variable
-
-try running
 
+#### Clone this github repository
 ```
-python --version
+git clone https://github.com/lukas/ml-class.git
+cd ml-class
 ```
 
-You should see something like
+#### libraries
 
 ```
-Python 3.6.1 :: Anaconda 4.4.0 (64-bit)
+pip install wandb
+conda install -c conda-forge scikit-learn
+conda install -c conda-forge tensorflow
+conda install -c conda-forge keras
 ```
 
+### Linux and Mac OS X
+#### Install python
+
+You can download python from https://www.python.org/downloads/.  There are more detailed instructions for windows installation at https://www.howtogeek.com/197947/how-to-install-python-on-windows/.  
+
+The material should work with python 2 or 3.  *On Windows, you need to install thre 64 bit version of python 3.5 or 3.6 in order to install tensorflow*.
+
 #### Clone this github repository
 ```
 git clone https://github.com/lukas/ml-class.git
 cd ml-class
 ```
 
-#### libraries
+If you get an error message here, most likely you don't have git installed.  Go to https://www.atlassian.com/git/tutorials/install-git for intructions on installing git.
 
+#### Install necessary pip libraries
 ```
-conda install -c conda-forge scikit-learn
-conda install -c conda-forge tensorflow
-conda install -c conda-forge keras
+pip install wandb
+pip install pandas
+pip install scikit-learn
+pip install tensorflow
+pip install keras
+```
+
+#### Install python libraries for optional material
+```
+pip install h5py
+pip install flask
+pip install scikit-image
+pip install scipy
+pip install pillow
 ```
 
 
@@ -192,16 +202,16 @@ Order of presentation of files, if you want to follow along
 - keras-scikit-learn.py
 - keras-one-hot.py
 - log-loss.py
-- keras-perceptron-1.py
-- keras-perceptron-2.py
-- keras-perceptron-3.py
-- keras-perceptron-4.py
-- keras-perceptron-checkpoint.py
-- keras-perceptron-save.py
-- keras-perceptron-load.py
-- keras-perceptron-regression.py
-- keras-mlp.py
-- keras-dropout.py
+- keras-perceptron/perceptron-1.py
+- keras-perceptron/perceptron-2.py
+- keras-perceptron/perceptron-3.py
+- keras-perceptron/perceptron-4.py
+- keras-perceptron/perceptron-checkpoint.py
+- keras-perceptron/perceptron-save.py
+- keras-perceptron/perceptron-load.py
+- keras-perceptron/perceptron-regression.py
+- keras-mlp/mlp.py
+- keras-mlp/dropout.py
 
 ### tensorflow
 - tensorflow-mult.py
@@ -214,9 +224,9 @@ Order of presentation of files, if you want to follow along
 ### conv neural nets
 - convolution-demo.py
 - maxpool-demo.py
-- keras-cnn-1.py
-- keras-cnn-2.py
-- keras-cnn-inspect.py
+- keras-cnn/cnn-1.py
+- keras-cnn/cnn-2.py
+- keras-cnn/cnn-inspect.py
 
 ### deep dream
 - keras-deep-dream.py

keras-cnn-wandb.py → keras-cnn/cnn-1.py

@@ -8,18 +8,8 @@
 run = wandb.init()
 config = run.config
 
-
 (X_train, y_train), (X_test, y_test) = mnist.load_data()
 
-config = wandb.run.config
-config.first_layer_convs = 32
-config.first_layer_conv_width = 3
-config.first_layer_conv_height = 3
-config.dropout = 0.2
-config.dense_layer_size = 128
-config.img_width=28
-config.img_height=28
-
 X_train = X_train.astype('float32')
 X_train /= 255.
 X_test = X_test.astype('float32')
@@ -29,14 +19,12 @@
 X_train = X_train.reshape(X_train.shape[0], config.img_width, config.img_height, 1)
 X_test = X_test.reshape(X_test.shape[0], config.img_width, config.img_height, 1)
 
-
 # one hot encode outputs
 y_train = np_utils.to_categorical(y_train)
 y_test = np_utils.to_categorical(y_test)
 num_classes = y_test.shape[1]
 
 # build model
-
 model = Sequential()
 model.add(Conv2D(32,
     (config.first_layer_conv_width, config.first_layer_conv_height),
@@ -48,5 +36,5 @@
 
 model.compile(loss='categorical_crossentropy', optimizer='adam',
                 metrics=['accuracy'])
-model.fit(X_train, y_train,  validation_data=(X_test, y_test),
-                                callbacks=[WandbKerasCallback()], epochs=config.epochs)
+model.fit(X_train, y_train, validation_data=(X_test, y_test),
+                            callbacks=[WandbKerasCallback()], epochs=config.epochs)

keras-cnn-1.py → keras-cnn/cnn-2.py

@@ -2,42 +2,41 @@
 from keras.models import Sequential
 from keras.layers import Conv2D, MaxPooling2D, Dropout, Dense, Flatten
 from keras.utils import np_utils
-
 from wandb.wandb_keras import WandbKerasCallback
 import wandb
 
 run = wandb.init()
 config = run.config
 
-
 (X_train, y_train), (X_test, y_test) = mnist.load_data()
 
-img_width=28
-img_height=28
-
 X_train = X_train.astype('float32')
 X_train /= 255.
 X_test = X_test.astype('float32')
 X_test /= 255.
 
 #reshape input data
-X_train = X_train.reshape(X_train.shape[0], img_width, img_height, 1)
-X_test = X_test.reshape(X_test.shape[0], img_width, img_height, 1)
+X_train = X_train.reshape(X_train.shape[0], config.img_width, config.img_height, 1)
+X_test = X_test.reshape(X_test.shape[0], config.img_width, config.img_height, 1)
 
 # one hot encode outputs
 y_train = np_utils.to_categorical(y_train)
 y_test = np_utils.to_categorical(y_test)
 num_classes = y_test.shape[1]
 
-# build model
-
 model = Sequential()
-model.add(Conv2D(32, (3, 3), input_shape=(img_width, img_height,1), activation='relu'))
+model.add(Conv2D(8, (5, 5), input_shape=(config.img_width, config.img_height,1), activation='relu'))
+model.add(MaxPooling2D(pool_size=(2, 2)))
+model.add(Conv2D(16, (5, 5), input_shape=(config.img_width, config.img_height,1), activation='relu'))
 model.add(MaxPooling2D(pool_size=(2, 2)))
 model.add(Dropout(0.2))
 model.add(Flatten())
-model.add(Dense(128, activation='relu'))
+model.add(Dense(32, activation='relu'))
 model.add(Dense(num_classes, activation='softmax'))
 
-model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
-model.fit(X_train, y_train)
+model.compile(loss='categorical_crossentropy', optimizer='adam',
+                metrics=['accuracy'])
+model.fit(X_train, y_train, validation_data=(X_test, y_test),
+                            callbacks=[WandbKerasCallback()], epochs=config.epochs)
+
+

keras-cnn/config-defaults.yaml

@@ -0,0 +1,26 @@
+wandb_version: 1
+
+first_layer_convs:
+   desc: Number of convolutions in the first layer
+   value: 32
+first_layer_conv_width:
+   desc: The width of the convolution
+   value: 3
+first_layer_conv_height:
+   desc: The height of the convolution
+   value: 3
+dropout:
+   desc: Percentage of neurons to dropout
+   value: 0.2
+dense_layer_size:
+   desc: The size of the dense layer
+   value: 128
+img_width:
+   desc: The width of the input image
+   value: 28
+img_height:
+   desc: The height of the input image
+   value: 28
+epochs:
+   desc: The number of epochs to run
+   value: 10

keras-cnn/wandb/settings

@@ -0,0 +1,3 @@
+[default]
+entity: qualcomm
+project: cnn

keras-mlp/config-defaults.yaml

@@ -0,0 +1,11 @@
+wandb_version: 1
+
+# Example variables below. Uncomment (remove leading '# ') to use them, or just
+# delete and create your own.
+
+epochs:
+  desc: Number of epochs to train over
+  value: 10
+hidden_nodes:
+  desc: Size of the hidden layer
+  value: 100