from mr.learn.unsupervised.lca import Lca from mr.learn.supervised.perceptron import Perceptron ## Load images nload = 70000 #train, test, vp = MnistDataset(nload).split(nload * 1 // 7) train, test, vp = MnistDataset(nload).split(10000) ## Set up model print ("Building model") model = Scaffold() c = ConvolveLayer(layer = Lca(15), visualParams = vp, convSize = 7, convStride = 3) c._init(len(train[0][0]), None) model.layer(c) #p = PoolLayer(visualParams = c.visualParams) #model.layer(p) model.layer(Perceptron()) ## Train and test model print ("Training model") model.fit(*train) print (model.layers[0].nOutputs) print (model.layers[0].nOutputsConvolved) path = 'visualize.png' path2 = 'visualize1.png' model.visualize(vp, path) model.visualize(vp, path2, inputs = test[0][0]) print ("Testing model") print (model.score(*test))
c = ConvolveLayer(layer = Lca(15), visualParams = vp, convSize = cnn_params[0], convStride = cnn_params[1]) c.init(len(train[0][0]), None) model.layer(c) p = PoolLayer(visualParams = c.visualParams) model.layer(p) model.layer(Perceptron()) ## Train and test model print ("Training model") start = datetime.datetime.now() model.fit(*train) print (model.layers[0].nOutputs) print (model.layers[0].nOutputsConvolved) ''' path = 'visualize.png' path2 = 'visualize2.png' model.visualize(vp, path) model.visualize(vp, path2, inputs = test[0][0]) ''' stop = datetime.datetime.now() train_min = (stop - start).total_seconds() / 60 print ("Total min to train: {}".format(train_min)) print ("Testing model") start = datetime.datetime.now() model2 = TowerScaffold() xP = model.predict(test[0], False) xP[xP > 0.5] = 1