Premier League stats & predictions

Squadta E vs Squadrta F

• Falco destro della Squadrta E**: Noto per le sue accelerazioni fulminee e cross precisi, è uno dei giocatori più temuti sulle fasce del campionato.
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
   zhangzhe2016/CS231n<|file_sep|>/README.md # CS231n Stanford's CS231n: Convolutional Neural Networks for Visual Recognition Course ## Lecture Slides [CS231n lecture notes](http://cs231n.github.io/) ## Assignments ### Assignment I [Assignment I](https://github.com/zhangzhe2016/CS231n/blob/master/assignment1/assignment1/cs231n/assignment1.ipynb) ### Assignment II [Assignment II](https://github.com/zhangzhe2016/CS231n/blob/master/assignment2/cs231n/assignment2.ipynb) ### Assignment III [Assignment III](https://github.com/zhangzhe2016/CS231n/blob/master/assignment3/cs231n/assignment3.ipynb) <|repo_name|>zhangzhe2016/CS231n<|file_sep|>/assignment2/cs231n/classifiers/rnn.py import numpy as np from cs231n.layers import * from cs231n.layer_utils import * class RNN(object): """ RNN model. """ def __init__(self, input_dim=28*28, hidden_dim=100, wordvec_dim=32, num_classes=10, bptt_truncate=-1): """ Initialize the RNN. We assume an input_dim word vector dimension (each word will be represented by a vector with this dimension). We also assume a fixed number of hidden units. Inputs: - input_dim: Dimension of word vectors. - hidden_dim: Number of hidden units. - wordvec_dim: Dimension of word vectors. - num_classes: Number of classes. """ self.input_dim = input_dim self.wordvec_dim = wordvec_dim self.hidden_dim = hidden_dim self.num_classes = num_classes self.bptt_truncate = bptt_truncate # Initialize the parameters of the model. self.U = np.random.uniform(-np.sqrt(0.01), np.sqrt(0.01), (hidden_dim, wordvec_dim)) self.V = np.random.uniform(-np.sqrt(0.01), np.sqrt(0.01), (num_classes, hidden_dim)) self.W = np.random.uniform(-np.sqrt(0.01), np.sqrt(0.01), (hidden_dim, hidden_dim)) def loss(self, X, Y): """ Compute loss and gradient for the entire sequence of data. Inputs: - X: Input data for the entire timeseries (list length T with each element shape (N,D)) where D is the dimension of the data. In our case D should be equal to 'wordvec_dim'. - Y: Labels for the entire timeseries (list length T where each element is an integer giving labels). In our case labels are between 0 and 'num_classes'. Returns: - loss as single float - gradients with respect to self.U, self.V and self.W stored in dictionary grads """ N = X[0].shape[0] T = len(X) # Perform forward propagation through time to compute the loss at every timestep. # Remember to perform parameter caching at every timestep by storing h_t and z_t. # You can use the rnn_step_forward function defined above. # Store all hidden states and inputs at every timestep using lists hs and zs. # Make sure you set the first element of hs to be zero before starting the loop! # Compute the total loss by summing up loss at every timestep. # Perform backpropagation through time to compute gradients at every timestep # Remember to truncate backpropagation after some number of steps by using bptt_truncate # You can use rnn_step_backward function defined above. # Store all gradients at every timestep using lists dU,dV,dW # Update gradients dU,dV,dW by summing up gradients at every timestep # Clip your gradients to mitigate exploding gradients # Return the loss and gradients <|file_sep|># CS231N - Convolutional Neural Networks for Visual Recognition Stanford University Fall Quarter ## Week I ### Lecture I #### Neural Networks Basics ##### Machine Learning Paradigms **Supervised learning:** Given labeled examples $(x,y)$ learn to predict $y$ from $x$. **Unsupervised learning:** Given unlabeled examples $x$ learn some structure about $x$. **Reinforcement learning:** Given an environment learn how to act within that environment. **Deep learning:** Model complex functions by composing simple functions together. **Why deep?** - Compositional nature of visual concepts allows us to represent high-level concepts as compositions of low-level concepts. ##### Neural Network Models Neural networks are composed of layers that transform their inputs $x$ into some output $y$. The network computes $y=f(x; theta)$ where $theta$ is the parameter vector. Neural networks are composed of layers that transform their inputs $x^{(l)}$ into some output $x^{(l+1)}$. The network computes $x^{(L)}=f(x^{(0)}; theta)$ where $theta$ is the parameter vector. A neural network layer is defined as: $$ x^{(l+1)} = f(x^{(l)}, W^{(l)}, b^{(l)}) $$ where $W^{(l)}$ and $b^{(l)}$ are learned parameters. The following are two examples: **Linear layer** $$ x^{(l+1)}=W^{(l)}x^{(l)} + b^{(l)} $$ **Logistic regression layer** $$ x^{(l+1)}=sigma(W^{(l)}x^{(l)} + b^{(l)}) $$ **Activation function** A function that introduces nonlinearity into neural network models so that we can represent more complex functions. **Loss function** Measures how good our model is at predicting some desired output given some input. Example: $$ L(hat{y}, y)=-sum_i y_ilog hat{y}_i $$ where $hat{y}=sigma(z)$ and $z=W^Tx+b$ ##### Training Neural Networks via Gradient Descent The goal of training is to minimize our loss function over our dataset by adjusting our parameters $theta$. $$ min_theta