Liga de Tineret East stats & predictions

Risorse Aggiuntive: Espandi la Tua Comprensione del Calcio!

Oltre agli aggiornamenti giornalieri sui match e alle previsioni, offriamo una varietà di risorse educative per aiutarti ad approfondire ulteriormente la tua comprensione del mondo del calcio. Che tu sia interessato alla storia dello sport o alle sue tecniche avanzate, abbiamo qualcosa da offrire!

• Corsi Online Gratuiti: Iscriviti ai nostri corsi gratuiti su vari aspetti del gioco – dall'analisi tattica all'influenza sociale del calcio.
• Biblioteca Digitale: Accesso a una vasta gamma di libri digitalizzati su storie famose, biografie dei giocatori ed evoluzione del gioco nel tempo.rpgoldstein/Convolutional_Networks<|file_sep|>/src/generate_data.py import numpy as np from scipy import misc import os from PIL import Image from sklearn.preprocessing import OneHotEncoder def generate_data(input_file_path): #input_file_path = 'data/train/00000.jpg' #img = misc.imread(input_file_path) img = Image.open(input_file_path) img = np.array(img) x = [] y = [] for row in range(0,img.shape[0]-24+1): for col in range(0,img.shape[1]-24+1): patch = img[row:(row+24),col:(col+24),:] x.append(patch.reshape(24*24*3)) y.append(img[row+12,col+12,:]) x = np.array(x) y = np.array(y) return x,y def generate_one_hot_labels(y): encoder = OneHotEncoder() y = y.reshape(y.shape[0],1) encoder.fit(y) y_one_hot = encoder.transform(y).toarray() return y_one_hot def generate_data_batch(input_dir,output_dir): batch_size = int(os.listdir(input_dir)[0].split('.')[0]) x = [] y = [] for filename in os.listdir(input_dir): if not filename.endswith('.npy'): continue if int(filename.split('.')[0]) != batch_size: continue data_x = np.load(os.path.join(input_dir,filename)) for row in range(0,data_x.shape[0]): if row %10000 ==0: print('row : '+str(row)+'/'+str(data_x.shape[0])) patch = data_x[row,:,:,:] x.append(patch.reshape(24*24*3)) label_row_num = int(row/(data_x.shape[1]-24+1)) label_col_num = row%(data_x.shape[1]-24+1) label_row_num += int(data_x.shape[1]/2-12) label_col_num += int(data_x.shape[2]/2-12) label_patch_row_num = int(label_row_num/24) label_patch_col_num = int(label_col_num/24) label_row_offset_in_patch = label_row_num%24 label_col_offset_in_patch = label_col_num%24 label_patch_row_num *= (data_x.shape[1]/24)-1 label_patch_col_num *= (data_x.shape[2]/24)-1 final_label_row_num = label_patch_row_num + label_row_offset_in_patch final_label_col_num = label_patch_col_num + label_col_offset_in_patch y.append(data_x[label_patch_row_num,label_patch_col_num,label_row_offset_in_patch,label_col_offset_in_patch,:]) batch_size +=1 if batch_size ==100: break x = np.array(x) print(x.shape) np.save(os.path.join(output_dir,'x_batch_'+str(batch_size)+'.npy'),x) #generate_data_batch('data/train','data/train_batches') #generate_data_batch('data/test','data/test_batches') def main(): # input_dir = 'data/train' # output_dir ='data/train_batches' # # generate_data_batch(input_dir,output_dir) # # input_dir = 'data/test' # output_dir ='data/test_batches' # # generate_data_batch(input_dir,output_dir) # input_file_path ='C:/Users/Roy Goldstein/Desktop/Spring_2017/Machine_Learning/Project/data/train/00001.jpg' # x,y= generate_data(input_file_path) # # print(x.shape,y.shape) if __name__=='__main__': main() <|file_sep|># Convolutional Networks Convolutional neural networks (CNN) are used to learn image classification tasks by training on images of flowers and using the learned features to classify new flower images. The CNN was trained on a large number of images of flowers from the following classes: * Daisy * Dandelion * Rose * Sunflower * Tulip Each image is an RGB image with a size of approximately [512 x 512 x 3]. ## Features of the CNN The CNN consists of four convolutional layers followed by two fully connected layers and one softmax layer. ### Convolutional Layers Each convolutional layer consists of several convolution filters that are convolved across the input image and produce an output feature map. The number of filters used for each layer is as follows: Layer | Number of Filters ----- | ----------------- 1 | 32 2 | 64 3 |128 4 |256 Each filter has a size of [5 x 5]. After each convolution layer is a max pooling layer with pooling size [2 x 2] and stride [2 x 2]. This reduces the size of the feature map by half. ### Fully Connected Layers After all the convolutional layers is two fully connected layers with [1024] units each. The output of the second fully connected layer is fed into a softmax layer with five output nodes. These nodes correspond to the probability that the image belongs to one of the five classes. ### Backpropagation Backpropagation is used to train the network. The network is trained on patches of images taken from the full image. This allows for more training examples which speeds up training time. Since each patch comes from an image that belongs to one class we can easily assign a one-hot encoded vector to each patch. Each patch is assigned the same vector as its parent image. ## Training and Testing The network was trained on patches taken from images found in "data/train". Testing was done on patches taken from images found in "data/test". Each epoch took approximately three hours to run. ## Results ### Training Accuracy  ### Testing Accuracy  ### Classification Accuracy Per Class  ### Confusion Matrix <|repo_name|>rpgoldstein/Convolutional_Networks<|file_sep|>/src/convnet.py import numpy as np import matplotlib.pyplot as plt import timeit from scipy.signal import convolve2d as convolve_2d class convnet(): def __init__(self): self.input_shape =(None,None,None,None) #x,y,z,batch self.conv_filter_shape_1=(32,5,5,self.input_shape[-1]) #num_filters,x,y,z self.conv_filter_shape_2=(64,self.conv_filter_shape_1[0],5,5) #num_filters,x,y,z self.conv_filter_shape_3=(128,self.conv_filter_shape_2[0],5,5) #num_filters,x,y,z self.conv_filter_shape_4=(256,self.conv_filter_shape_3[0],5,5) #num_filters,x,y,z self.fc_layer_shape_1=(1024,self._get_conv_out_size(self.conv_filter_shape_4)) #units,in_features self.fc_layer_shape_2=(self.fc_layer_shape_1[0],self.fc_layer_shape_1[0]) #units,in_features self.num_classes=5 self.learning_rate=10**-5 self.num_epochs=20 self.batch_size=100 self.momentum=0.9 self._initialize_weights() print(self.w_conv_1.shape) #