original: https://gist.github.com/simonw/50e14b9a3e829355d6d43f0f12f91e74
I started a project on a Hobby Dev plan (free, limit 10,000 rows), and then later needed to upgrade it to Hobby Basic ($9/month, limit 10,000,000 rows).
<appname> with the name of your Heroku app.
| git clone https://github.com/rescience-c/template.git | |
| cd template | |
| brew tap caskroom/cask | |
| brew cask install mactex | |
| make draft |
| rm -rf `find -type d -name .ipynb_checkpoints` |
| # Inspiration from | |
| # https://uoa-eresearch.github.io/eresearch-cookbook/recipe/2014/11/20/conda/ | |
| # and | |
| # https://stackoverflow.com/questions/39604271/conda-environments-not-showing-up-in-jupyter-notebook | |
| # Check if conda is up to date | |
| conda update conda | |
| # Create a virtual environment for your project | |
| conda create -n myenv python=x.x anaconda |
| import cv2 | |
| def imgread(path): | |
| print "Image:", path.split("/")[-1] | |
| # Read in the image using python opencv | |
| img = cv2.imread(path) | |
| img = img / 255.0 | |
| print "Raw Image Shape: ", img.shape | |
| # Center crop the image |
| @ops.RegisterGradient("GuidedRelu") | |
| def _GuidedReluGrad(op, grad): | |
| return tf.where(0. < grad, gen_nn_ops._relu_grad(grad, op.outputs[0]), tf.zeros(grad.get_shape())) |
| # Number of iterations to run. | |
| ITERATIONS = 500 | |
| sess.run(tf.global_variables_initializer()) | |
| input_image = np.reshape(gen_image, ((1,) + gen_image.shape)) | |
| sess.run(model['input'].assign(input_image)) | |
| for it in range(ITERATIONS): | |
| sess.run(train_step) | |
| if it%50 == 0: | |
| # Print every 50 iteration. |
| # Content loss emphasis. | |
| ALPHA = 0.0025 | |
| # Style loss emphasis. | |
| BETA = 1 | |
| total_loss = ALPHA * content_loss + BETA * style_loss | |
| optimizer = tf.train.AdamOptimizer(2.0) | |
| train_step = optimizer.minimize(total_loss) |
| sess = tf.InteractiveSession() | |
| model = load_vgg_model(VGG_MODEL) | |
| # Construct content_loss using content_image. | |
| content_image_list = np.reshape(content_image, ((1,) + content_image.shape)) | |
| sess.run(model['input'].assign(content_image_list)) | |
| content_loss = content_loss_func(sess, model) | |
| # Construct style_loss using style_image. |
| STYLE_LAYERS = [ | |
| ('conv1_1', 0.5), | |
| ('conv2_1', 0.5), | |
| ('conv3_1', 0.5), | |
| ('conv4_1', 0.5), | |
| ('conv5_1', 0.5), | |
| ] | |
| def style_loss_func(sess, model): | |
| def _gram_matrix(feat): |
