training neural network ok, needs too much ram. Started implementation of the server

nnTrain/.gitignore

@@ -0,0 +1 @@
+trainBACKUP.py

nnTrain/README.md

@@ -0,0 +1,24 @@
+# serverImgPredictor
+Need the file dataset.data
+
+### install Flask
+http://flask.pocoo.org/docs/0.12/quickstart/#a-minimal-application
+(sudo) pip install Flask
+
+pip install flask_restful
+pip install flask-jsonpify
+
+### install scikit-neuralnetwork
+https://scikit-neuralnetwork.readthedocs.io/en/latest/guide_installation.html
+pip install scikit-neuralnetwork
+
+also need to upgrade the Lasagne library:
+(sudo) pip install --upgrade https://github.com/Lasagne/Lasagne/archive/master.zip
+
+
+## Run
+python train.py
+
+will generate nn.pkl
+
+copy nn.pkl to the serverPredictor directory

nnTrain/predict.py

@@ -0,0 +1,32 @@
+from sklearn.neural_network import MLPClassifier
+from skimage import io
+
+img1 = io.imread("imgs/25.png")
+img2 = io.imread("imgs/24.png")
+img3 = io.imread("imgs/104.png")
+
+img4 = io.imread("otherimgs/image_0008.jpg")
+
+
+data_train = [img1, img2, img3, img4]
+data_labels = [1, 1, 1, 0]
+data_test = [img4, img3]
+clf = MLPClassifier(solver='lbfgs', alpha=1e-5,
+                    hidden_layer_sizes=(5,2), random_state=1)
+clf.fit(data_train, data_labels)
+
+clf.predict(data_test)
+
+print "MPLClassifier values:"
+[coef.shape for coef in clf.coefs_]
+
+
+
+'''
+images_and_predictions = list(zip(digits.images[n_samples // 2:], predicted))
+for index, (image, prediction) in enumerate(images_and_predictions[:4]):
+    plt.subplot(2, 4, index + 5)
+    plt.axis('off')
+    plt.imshow(image, cmap=plt.cm.gray_r, interpolation='nearest')
+    plt.title('Prediction: %i' % prediction)
+'''

nnTrain/train.py

@@ -0,0 +1,62 @@
+import matplotlib.pyplot as plt
+import numpy as np
+from random import randint
+import pickle
+from sknn.mlp import Classifier, Layer, Convolution
+
+def datasetToTrainAndTestData(dataset, numtest):
+    np.random.shuffle(dataset)
+    print "length total data:" + str(len(dataset))
+
+    traindata = np.copy(dataset)
+    testdata = []
+    for i in range(numtest):
+        #get random integer between 0 and the total amount of images in the dataset
+        n = randint(0, len(traindata))
+        testdata.append(dataset[n])
+
+        #delete the n image (dataset[n]) of the traindata
+        traindata = np.delete(traindata, n, axis=0)
+    testdataNP = np.array(testdata)
+    return traindata, testdataNP
+
+
+#read the dataset made with the 'imagesToDataset' repository
+dataset = np.load('dataset.npy')
+
+traindata, testdata = datasetToTrainAndTestData(dataset, 10)
+print "length traindata: " + str(len(traindata))
+print "length testdata: " + str(len(testdata))
+
+#traindataAttributes contains all the pixels of each image
+traindataAttributes = traindata[:,0]
+traindataAttributes = np.array([[row] for row in traindataAttributes])
+
+#traindataLabels contains each label of each image
+traindataLabels = traindata[:,1]
+traindataLabels = traindataLabels.astype('int')
+
+#testdataAttributes contains the pixels of the test images
+testdataAttributes = testdata[:,0]
+testdataAttributes = np.array([[row] for row in testdataAttributes])
+
+#testdataLabels contains each label of each image
+testdataLabels = testdata[:,1]
+testdataLabels = testdataLabels.astype('int')
+
+#default: units=100, learning_rate=0.001, n_iter=25
+nn = Classifier(
+    layers=[
+        Layer("Sigmoid", units=10),
+        Layer("Softmax")],
+    learning_rate=0.001,
+    n_iter=20,
+    verbose=True)
+
+nn.fit(traindataAttributes, traindataLabels)
+
+print('\nTRAIN SCORE', nn.score(traindataAttributes, traindataLabels))
+print('TEST SCORE', nn.score(testdataAttributes, testdataLabels))
+
+#save the neural network configuration
+pickle.dump(nn, open('nn.pkl', 'wb'))