aaaa
/
learning_algorithms
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added ml experiments / projects

Browse Source
master
Alex 6 years ago
parent 169d99a82f
commit 5a33b132a8
11 changed files with 2009 additions and 0 deletions
Show Stats Download Patch File Download Diff File

102
DominantColorExtraction/getdominantcolor.py
Unescape Escape View File

@ -0,0 +1,102 @@
import cv2
import numpy as np
import matplotlib.pyplot as plt
from sklearn.cluster import KMeans
from os import listdir
import imageio
from PIL import Image, ImageDraw
def list_files(directory, extension):
return (f for f in listdir(directory) if f.endswith('.jpg') or f.endswith('.png'))
def find_histogram(clt):
"""
create a histogram with k clusters
:param: clt
:return:hist
"""
numLabels = np.arange(0, len(np.unique(clt.labels_)) + 1)
(hist, _) = np.histogram(clt.labels_, bins=numLabels)
hist = hist.astype("float")
hist /= hist.sum()
return hist
def plot_colors2(hist, centroids):
bar = np.zeros((50, 300, 3), dtype="uint8")
startX = 0
for (percent, color) in zip(hist, centroids):
# plot the relative percentage of each cluster
endX = startX + (percent * 300)
cv2.rectangle(bar, (int(startX), 0), (int(endX), 50),
color.astype("uint8").tolist(), -1)
startX = endX
# return the bar chart
return bar
def generateColorImage(img_input):
img = cv2.imread("images/"+img_input)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img_old = img
height=img.shape[0]
width=img.shape[1]
img = img.reshape((img.shape[0] * img.shape[1],3)) #represent as row*column,channel number
clt = KMeans(n_clusters=3) #cluster number
clt.fit(img)
hist = find_histogram(clt)
#print(hist, clt.cluster_centers_)
# size of image
canvas = (width, height)
# scale ration
scale = 1
thumb = canvas[0]/scale, canvas[1]/scale
# rectangles (width, height, left position, top position)
#print(width*hist[0])
frames = [(0, 0, 115, height), (width*hist[0], height, width*hist[0]+width,2), (100, 205, 120,200)]
# init canvas
im = Image.new('RGB', canvas, (255, 255, 255))
draw = ImageDraw.Draw(im)
draw.rectangle([0, 0, width*hist[0], height], fill=(int(clt.cluster_centers_[0][0]), int(clt.cluster_centers_[0][1]), int(clt.cluster_centers_[0][2])))
draw.rectangle([width*hist[0], 0, width*hist[0]+width*hist[1], height], fill=(int(clt.cluster_centers_[1][0]),int(clt.cluster_centers_[1][1]),int(clt.cluster_centers_[1][2])))
draw.rectangle([width*hist[0]+width*hist[1], 0, width*hist[0]+width*hist[1]+width*hist[2], height], fill=(int(clt.cluster_centers_[2][0]),int(clt.cluster_centers_[2][1]),int(clt.cluster_centers_[2][2])))
# make thumbnail
im.thumbnail(thumb)
# save image
im.save("output/"+img_input)
# images = []
# images.append(imageio.imread("images/"+img_input))
# images.append(imageio.imread("img/"+img_input))
# imageio.mimsave('gifs/'+img_input+".gif", images)
directory = 'images'
files = list_files(directory, "jpg")
for f in files:
print(f)
generateColorImage(f)
#bar = plot_colors2(hist, clt.cluster_centers_)
#plt.axis("off")
#plt.imshow(bar)
#plt.show()

58
ImageClassificationPython/predict.py
Unescape Escape View File

@ -0,0 +1,58 @@
# Part 3 - Making new predictions
import numpy as np
from PIL import Image, ImageDraw, ImageFont
from keras.preprocessing import image
from keras.models import model_from_yaml
from keras.preprocessing.image import ImageDataGenerator
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('images', nargs="*", help="images to classify")
args = parser.parse_args()
# load YAML and create model
yaml_file = open('model.yaml', 'r')
loaded_model_yaml = yaml_file.read()
yaml_file.close()
classifier = model_from_yaml(loaded_model_yaml)
# load weights into new model
classifier.load_weights("model.h5")
print("Loaded model from disk")
for f in args.images:
from keras.preprocessing import image
test_image = image.load_img(f, target_size = (64, 64))
test_image = image.img_to_array(test_image)
test_image = np.expand_dims(test_image, axis = 0)
result = classifier.predict(test_image)
print(result)
#WHAT ARE YOUR CLASSES?
if result[0][0] == 1:
prediction = 'rect'
else:
prediction = 'circle'
print("PREDICTION: {}".format(prediction))
#WRITE RESULT TO IMAGE
image = Image.open(f)
width, height = image.size
size = (width, height+100)
layer = Image.new('RGB', size, (255,255,255))
layer.paste(image, (0,0))
draw = ImageDraw.Draw(layer)
font = ImageFont.truetype('Roboto-Regular.ttf', size=45)
(x, y) = (50, height+20)
message = prediction
color = 'rgb(0, 0, 0)' # black color
draw.text((x, y), message, fill=color, font=font)
layer.save("{}.predicted.png".format(f))

11
ImageClassificationPython/requirements.txt
Unescape Escape View File

@ -0,0 +1,11 @@
h5py==2.8.0
Keras==2.2.4
Keras-Applications==1.0.6
Keras-Preprocessing==1.0.5
numpy==1.15.2
Pillow==5.3.0
protobuf==3.6.1
PyYAML==3.13
scipy==1.1.0
six==1.11.0
tensorflow==1.0.0

78
ImageClassificationPython/train.py
Unescape Escape View File

@ -0,0 +1,78 @@
# Convolutional Neural Network
# Installing Theano
# pip install --upgrade --no-deps git+git://github.com/Theano/Theano.git
# Installing Tensorflow
# pip install tensorflow
# Installing Keras
# pip install --upgrade keras
# Part 1 - Building the CNN
# Importing the Keras libraries and packages
from keras.models import Sequential
from keras.layers import Conv2D
from keras.layers import MaxPooling2D
from keras.layers import Flatten
from keras.layers import Dense
# Initialising the CNN
classifier = Sequential()
# Step 1 - Convolution
classifier.add(Conv2D(32, (3, 3), input_shape = (64, 64, 3), activation = 'relu'))
# Step 2 - Pooling
classifier.add(MaxPooling2D(pool_size = (2, 2)))
# Adding a second convolutional layer
classifier.add(Conv2D(32, (3, 3), activation = 'relu'))
classifier.add(MaxPooling2D(pool_size = (2, 2)))
# Step 3 - Flattening
classifier.add(Flatten())
# Step 4 - Full connection
classifier.add(Dense(units = 128, activation = 'relu'))
classifier.add(Dense(units = 1, activation = 'sigmoid'))
# Compiling the CNN
classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])
# Part 2 - Fitting the CNN to the images
from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale = 1./255,
shear_range = 0.2,
zoom_range = 0.2,
horizontal_flip = True)
test_datagen = ImageDataGenerator(rescale = 1./255)
training_set = train_datagen.flow_from_directory('dataset/training_set',
target_size = (64, 64),
batch_size = 32,
class_mode = 'binary')
test_set = test_datagen.flow_from_directory('dataset/test_set',
target_size = (64, 64),
batch_size = 32,
class_mode = 'binary')
classifier.fit_generator(training_set,
steps_per_epoch = 500,
epochs = 1,
validation_data = test_set,
validation_steps = 100)
# serialize model to YAML
model_yaml = classifier.to_yaml()
with open("model.yaml", "w") as yaml_file:
yaml_file.write(model_yaml)
# serialize weights to HDF5
classifier.save_weights("model.h5")
print("Saved model to disk")

162
VanillaJavascriptSimpleNN/NeuralNetwork.js
Unescape Escape View File

@ -0,0 +1,162 @@
class NeuralNetwork
{
constructor(numInputs, numOutputs, numHiddenLayers, numNeuronsPerHiddenLayer)
{
this.numInputs = numInputs;
this.numOutputs = numOutputs;
this.numHiddenLayers = numHiddenLayers;
this.numNeuronsPerHiddenLayer = numNeuronsPerHiddenLayer;
this.bias = 0.0;
this.activationResponse = 1.0;
this.neuronLayers = [];
this.createNetwork();
}
createNetwork()
{
//create the layers of the network
if (this.numHiddenLayers > 0)
{
//create first hidden layer
var firstHiddenLayer = new NeuronLayer(this.numNeuronsPerHiddenLayer, this.numInputs);
this.neuronLayers.push(firstHiddenLayer);
for (var i=0; i<this.numHiddenLayers-1; ++i)
{
var newHiddenLayer = new NeuronLayer(this.numNeuronsPerHiddenLayer, this.numNeuronsPerHiddenLayer);
this.neuronLayers.push(newHiddenLayer);
}
//create output layer
var outputLayer = new NeuronLayer(this.numOutputs, this.numNeuronsPerHiddenLayer);
this.neuronLayers.push(outputLayer);
}
else
{
//create output layer
var outputLayer = new NeuronLayer(this.numOutputs, this.numInputs);
this.neuronLayers.push(outputLayer);
}
}
update(inputs)
{
var outputs = [];
var cWeight = 0;
// If the number of inputs supplied is incorrect...
if (inputs.length!=this.numInputs)
{
return outputs; // Return empty outputs
}
// Loop through all layers
var inputLayer = true;
for (var i=0; i < this.numHiddenLayers + 1; ++i)
{
var neuronLayer = this.neuronLayers[i];
if (!inputLayer)
{
inputs = [];
inputs = inputs.concat(outputs);
}
else
{
inputLayer = false;
}
outputs = [];
cWeight = 0;
// For each neuron sum the (inputs * corresponding weights).
// Throw the total at our sigmoid function to get the output.
for (var j=0; j < neuronLayer.neurons.length; ++j)
{
var neuron = neuronLayer.neurons[j];
var totalInput = 0;
// For each weight...
for (var k=0; k < neuron.numInputs ; ++k) // -1 ???
{
// Multiply it with the input.
totalInput += neuron.weights[k] *
inputs[cWeight];
console.log("cweight "+cWeight);
console.log("neuron weight "+neuron.weights[k]);
console.log("input: "+inputs[cWeight]);
console.log("total input "+totalInput);
cWeight++;
}
// Add in the bias (final weight)
// totalInput += neuron.weights[neuron.weights.length-1] * this.bias;
// We can store the outputs from each layer as we generate them.
// The combined activation is first filtered through the sigmoid function
outputs.push(this.sigmoid(totalInput, this.activationResponse));
cWeight = 0;
}
}
return outputs;
}
sigmoid(totalInput, activationResponse)
{
return ( 1 / ( 1 + Math.exp(-totalInput / activationResponse)));
}
getWeights()
{
var weights = [];
//for each layer
for (var i=0; i<this.numHiddenLayers + 1; ++i)
{
//for each neuron
for (var j=0; j<this.neuronLayers[i].neurons.length; ++j)
{
//for each weight
for (var k=0; k<this.neuronLayers[i].neurons[j].numInputs; ++k)
{
weights.push(this.neuronLayers[i].neurons[j].weights[k]);
}
}
}
return weights;
}
setWeights(weights)
{
var cWeight = 0;
//for each layer
for (var i=0; i<this.numHiddenLayers + 1; ++i)
{
//for each neuron
for (var j=0; j<this.neuronLayers[i].neurons.length; ++j)
{
//for each weight
for (var k=0; k<this.neuronLayers[i].neurons[j].numInputs; ++k)
{
this.neuronLayers[i].neurons[j].weights[k] = weights[cWeight++];
}
}
}
}
}

15
VanillaJavascriptSimpleNN/Neuron.js
Unescape Escape View File

@ -0,0 +1,15 @@
class Neuron
{
constructor(numInputs)
{
this.weights = [];
this.numInputs = numInputs;
for (var i=0; i<numInputs+1; ++i)
{
var newWeight = -1 + (Math.random()*2);
//var newWeight = 1;
this.weights.push(newWeight);
}
}
}

14
VanillaJavascriptSimpleNN/NeuronLayer.js
Unescape Escape View File

@ -0,0 +1,14 @@
class NeuronLayer
{
constructor(numNeuronsPerHiddenLayer, numInputs)
{
this.neurons = [];
for (var i = 0; i < numNeuronsPerHiddenLayer; ++i)
{
var newNeuron = new Neuron(numInputs);
this.neurons.push(newNeuron);
}
}
}

225
VanillaJavascriptSimpleNN/index.html
Unescape Escape View File

@ -0,0 +1,225 @@
<html>
<head>
<script src="NeuralNetwork.js"></script>
<script src="Neuron.js"></script>
<script src="NeuronLayer.js"></script>
<style>
body{
padding:0;
margin:0;
overflow:hidden;
font-family:Helvetica;
font-size:10px;
}
input{
padding:0;
}
p{
padding:0;
margin:0;
}
#input{
position:absolute;
top:100px;
left:10px;
}
#first_layer{
position:absolute;
top: 100px;
left: 200px;
}
#second_layer{
position:absolute;
top: 100px;
left: 400px;
}
#output{
position:absolute;
top: 100px;
left: 600px;
}
#network_container{
position:absolute;
top: 100px;
left: 600px;
margin-left:-400px;
margin-top:-250px;
}
</style>
</head>
<body>
<canvas id="myCanvas" width="100%" height="100%"></canvas>
<div id="network_container">
<div id="input">
<input type="range" min="0" max="100" value="1" class="i0" id="i0">
<p id="i0_output"></p>
<input type="range" min="0" max="100" value="10" class="i1" id="i1">
<p id="i1_output"></p>
</div>
<div id = "first_layer">
<input type="range" min="-400" max="400" value="10" class="w0" id="w0">
<p id="w0_output"></p>
<input type="range" min="-400" max="400" value="10" class="w1" id="w1">
<p id="w1_output"></p><br><br>
<input type="range" min="-400" max="400" value="10" class="w2" id="w2">
<p id="w2_output"></p>
<input type="range" min="-400" max="400" value="10" class="w3" id="w3">
<p id="w3_output"></p><br><br>
<input type="range" min="-400" max="400" value="10" class="w4" id="w4">
<p id="w4_output"></p>
<input type="range" min="-400" max="400" value="10" class="w5" id="w5">
<p id="w5_output"></p>
</div>
<div id = "second_layer">
<input type="range" min="-400" max="400" value="10" class="w6" id="w6">
<p id="w6_output"></p>
<input type="range" min="-400" max="400" value="10" class="w7" id="w7">
<p id="w7_output"></p>
<input type="range" min="-400" max="400" value="10" class="w8" id="w8">
<p id="w8_output"></p><br><br>
<input type="range" min="-400" max="400" value="10" class="w9" id="w9">
<p id="w9_output"></p>
<input type="range" min="-400" max="400" value="10" class="w10" id="w10">
<p id="w10_output"></p>
<input type="range" min="-400" max="400" value="10" class="w11" id="w11">
<p id="w11_output"></p>
</div>
<div id="output">
<input id="clickMe" type="button" value="update network" onclick="update();" />
<input id="random" type="button" value="random network" onclick="random();" />
<p id="network_output"></p>
</div>
</div>
<script>
var numInputs = 2;
var numOutputs = 2;
var numHiddenLayers = 1;
var numNeuronsPerHiddenLayer = 3;
var slider = [];
var output = [];
var neuralNetwork = new NeuralNetwork(numInputs, numOutputs, numHiddenLayers, numNeuronsPerHiddenLayer);
var input0 = document.getElementById("i0")
var input_display0 = document.getElementById("i0_output")
input_display0.innerHTML = input0.value/100;
input0.oninput = function() {
input_display0.innerHTML = this.value/100;
}
var input1 = document.getElementById("i1")
var input_display1 = document.getElementById("i1_output")
input_display1.innerHTML = input1.value/100;
input1.oninput = function() {
input_display1.innerHTML = this.value/100;
}
//slider
for (var i = 0; i<12; i++){
slider.push(document.getElementById("w"+i));
output.push(document.getElementById("w"+i+"_output"));
output[i].innerHTML = slider[i].value/100;
slider[i].oninput = function() {
output[slider.indexOf(this)].innerHTML = this.value/100;
updateCanvas()
}
}
document.addEventListener('mousemove', function(e) {
console.log("move")
var x = e.clientX / window.innerWidth;
var y = e.clientY / window.innerHeight;
input0.value=x*100;
input1.value=y*100;
input_display0.innerHTML = x;
input_display1.innerHTML = y;
update([x,y])
})
//CANVAS STUFF
var c = document.getElementById("myCanvas");
var ctx = c.getContext("2d");
ctx.canvas.width = window.innerWidth;
ctx.canvas.height = window.innerHeight;
var stepsize = 50;
for(var x = 0; x < window.innerWidth; x = x+stepsize){
for(var y = 0; y < window.innerHeight; y = y+stepsize){
var outputs = neuralNetwork.update([x/window.innerWidth,y/window.innerHeight]);
ctx.fillStyle = 'hsl(' + 360 * outputs[0] + ', 50%,'+outputs[1]*100 +'%)';
//ctx.fillStyle= "rgb(255,20,255);"
console.log(outputs[1]*255)
ctx.fillRect(x,y,stepsize,stepsize);
}
}
function updateCanvas(){
var stepsize = 50;
for(var x = 0; x < window.innerWidth; x = x+stepsize){
for(var y = 0; y < window.innerHeight; y = y+stepsize){
var outputs = neuralNetwork.update([x/window.innerWidth,y/window.innerHeight]);
ctx.fillStyle = 'hsl(' + 360 * outputs[0] + ', 90%,'+outputs[1]*100+'%)';
//ctx.fillStyle= "rgb(255,20,255);"
console.log(outputs[1]*255)
ctx.fillRect(x,y,stepsize,stepsize);
}
}
}
function update(inputs){
var weights = neuralNetwork.getWeights();
var newWeights = [];
for (var i=0; i < weights.length; i++) {
newWeights.push(slider[i].value/100);
//newWeights.push(1);
}
neuralNetwork.setWeights(newWeights);
//var inputs = [input0.value/100, input1.value/100];
var outputs = neuralNetwork.update(inputs);
// console.log(neuralNetwork.getWeights());
var output_display = document.getElementById("network_output")
output_display.innerHTML = "<br> "+outputs[0] + "<br><br> "+outputs [1];
document.getElementById('network_container').style.left=outputs[0]*window.innerWidth;
document.getElementById('network_container').style.top=outputs[1]*window.innerHeight;
}
random()
function random(inputs){
for (var i = 0; i<slider.length; i++){
slider[i].value = Math.random()*800-400;
output[slider.indexOf(slider[i])].innerHTML = slider[i].value/100;
}
updateCanvas();
}
</script>
</body>
</html>

198
VanillaJavascriptVisualisation/index.html
Unescape Escape View File

@ -0,0 +1,198 @@
<html>
<head>
<title>neural network art</title>
<meta charset="UTF-8">
<meta name="viewport" content="width=320, initial-scale=1.0, maximum-scale=1.0, user-scalable=0"/>
<meta name="apple-mobile-web-app-capable" content="yes" />
<meta http-equiv="X-UA-Compatible" content="IE=edge">
<meta name="mobile-web-app-capable" content="yes">
<meta name="description" content="ōtoro.net">
<meta name="author" content="hardmaru">
<!-- extra styles -->
<style>
head {
margin: 0;
padding: 0;
}
body {
margin: 0;
padding: 0;
}
p {
padding: 10px;
font-family: Courier, "Helvetica Neue",Helvetica,Arial,sans-serif;
font-weight: 100;
font-size: 0.6em;
}
textarea {
padding: 5;
font-family: Courier, "Helvetica Neue",Helvetica,Arial,sans-serif;
font-weight: 100;
font-size: 0.5em;
}
</style>
</head>
<body>
<div id="p5Container">
</div>
<!-- jQuery -->
<script src="http://ajax.googleapis.com/ajax/libs/jquery/1.8.3/jquery.min.js"></script>
<script src="lib/p5.min.js"></script>
<script src="lib/recurrent.js"></script>
<script>
// neural network random art generator
// settings
// actual size of generated image
var sizeh = 300;
var sizew = 300;
var sizeImage = sizeh*sizew;
var nH, nW, nImage;
var mask;
// settings of nnet:
var networkSize = 8;
var nHidden = 10;
var nOut = 3; // r, g, b layers
// support variables:
var img;
var img2;
var G = new R.Graph(false);
var initModel = function() {
"use strict";
var model = [];
var i;
var randomSize = 1.0;
// define the model below:
model.w_in = R.RandMat(networkSize, 3, 0, randomSize); // x, y, and bias
// model['w_0'] = R.RandMat(networkSize, networkSize, 0, randomSize);
for (i = 0; i < nHidden; i++) {
model['w_'+i] = R.RandMat(networkSize, networkSize, 0, randomSize);
}
model.w_out = R.RandMat(nOut, networkSize, 0, randomSize); // output layer
console.log(model)
return model;
};
var forwardNetwork = function(G, model, x_, y_) {
// x_, y_ is a normal javascript float, will be converted to a mat object below
// G is a graph to amend ops to
var x = new R.Mat(3, 1); // input
var i;
x.set(0, 0, x_);
x.set(1, 0, y_);
x.set(2, 0, 1.0); // bias.
var out;
out = G.tanh(G.mul(model.w_in, x));
for (i = 0; i < nHidden; i++) {
out = G.tanh(G.mul(model['w_'+i], out));
}
out = G.sigmoid(G.mul(model.w_out, out));
console.log(out)
return out;
};
function getColorAt(model, x, y) {
// function that returns a color given coordintes (x, y)
// (x, y) are scaled to -0.5 -> 0.5 for image recognition later
// but it can be behond the +/- 0.5 for generation above and beyond
// recognition limits
var r, g, b;
var out = forwardNetwork(G, model, x, y);
r = out.w[0]*255.0;
g = out.w[1]*255.0;
b = out.w[2]*255.0;
return color(r, g, b);
}
function genImage(img, model) {
var i, j, m, n;
img.loadPixels();
for (i = 0, m=img.width; i < m; i++) {
for (j = 0, n=img.height; j < n; j++) {
img.set(i, j, getColorAt(model, i/sizeh-0.5,j/sizew-0.5));
}
}
img.updatePixels();
}
function setup() {
"use strict";
var myCanvas;
myCanvas = createCanvas(windowWidth,windowHeight);
myCanvas.parent('p5Container');
nW = Math.max(Math.floor(windowWidth/sizew), 1);
nH = Math.max(Math.floor(windowHeight/sizeh), 1);
nImage = nH*nW;
mask = R.zeros(nImage);
//img.resize(320*1.0, 320*1.0);
//img.save('genart.png','png');
//noLoop();
img = createImage(sizeh, sizew);
frameRate(30);
}
function getRandomLocation() {
var i, result=0, r;
for (i=0;i<nImage;i++) {
result += mask[i];
}
if (result === nImage) {
mask = R.zeros(nImage);
}
do {
r = R.randi(0, nImage);
} while (mask[r] !== 0);
mask[r] = 1;
return r;
}
function displayImage(n) {
var row = Math.floor(n/nW);
var col = n % nW;
image(img, col*sizew, row*sizeh);
}
function draw() {
model = initModel();
genImage(img, model);
displayImage(getRandomLocation());
}
</script>
</body>
</html>

5
VanillaJavascriptVisualisation/lib/p5.min.js vendored
View File

File diff suppressed because one or more lines are too long

1141
VanillaJavascriptVisualisation/lib/recurrent.js
View File

File diff suppressed because it is too large Load Diff
Write Preview
Loading…
Cancel
Save