numpy复现算法lstm算法内含数据集以及教程

""" """ import numpy as np def softmax(x): e_x = np.exp(x - np.max(x)) return e_x / e_x.sum(axis=0) def sigmoid(x): return 1 / (1 + np.exp(-x)) class LSTM: def __init__(self, epochs=20, n_a=16, alpha=0.01, batch_size=32): """ :param epochs: 迭代次数 :param n_a: 隐藏层节点数 :param alpha: 梯度下降参数 :param batch_size: 每个batch大小 """ self.epochs = epochs self.n_a = n_a self.alpha = alpha self.parameters = {} self.loss = 0.0 self.n_x = 2 self.n_y = 2 self.m = batch_size def initialize_parameters(self, n_a, n_x, n_y): """ :param n_a: 每个cell输出a的维度 RNN单元的单元数 :param n_x: 每个cell输入xi的维度 :param n_y: 每个cell输出yi的维度 """ np.random.seed(1) Wf = np.random.randn(n_a, n_a + n_x) * 0.01 # randn函数返回一个或一组样本，具有标准正态分布。(50,50+27) bf = np.zeros((n_a, 1)) # (50,1) Wi = np.random.randn(n_a, n_a + n_x) * 0.01 # (50,50+27)=(50,77)【h(t-1),x(t)】=(50+27,1)=(77,1) bi = np.zeros((n_a, 1)) # (50,1) Wc = np.random.randn(n_a, n_a + n_x) * 0.01 # (50,50+27) 【h(t-1),x(t)】=(50+27,1)=(77,1) bc = np.zeros((n_a, 1)) # (50,1) Wo = np.random.randn(n_a, n_a + n_x) * 0.01 # (50,50+27) 【h(t-1),x(t)】=(50+27,1)=(77,1) bo = np.zeros((n_a, 1)) # (50,1) # Wo*【h(t-1),x(t)】+bo=(50,1) Wy = np.random.randn(n_y, n_a) * 0.01 # (27,50) h=(50,1) by = np.zeros((n_y, 1)) # (27,1) # Wy*h+by=(27,1) self.parameters = { "Wf": Wf, "bf": bf, "Wi": Wi, "bi": bi, "Wc": Wc, "bc": bc, "Wo": Wo, "bo": bo, "Wy": Wy, "by": by, } self.n_x = n_x # 27 self.n_y = n_y # 27 def lstm_cell_forward(self, xt, a_prev, c_prev): """ Implement a single forward step of the LSTM-cell as described in Figure (4) 如图（4）所示，执行LSTM单元的单个前进步骤 Arguments: xt -- your input data at timestep "t", numpy array of shape (n_x, m). 时间步 “t”处输入的数据，形状的numpy数组（n_x，m） a_prev -- Hidden state at timestep "t-1", numpy array of shape (n_a, m) 时间步“t-1”处的隐藏状态，形状的numpy数组（n_a，m） c_prev -- Memory state at timestep "t-1", numpy array of shape (n_a, m) 时间步“t-1”处的记忆状态，形状的numpy数组（n_a，m） parameters -- python dictionary containing: python字典包含： Wf -- Weight matrix of the forget gate, numpy array of shape (n_a, n_a + n_x)遗忘门的权重矩阵，形状的numpy数组 bf -- Bias of the forget gate, numpy array of shape (n_a, 1) 遗忘门的偏差，形状的numpy数组 Wi -- Weight matrix of the update gate, numpy array of shape (n_a, n_a + n_x)更新门的权重矩阵，形状的numpy数组 bi -- Bias of the update gate, numpy array of shape (n_a, 1) 更新门的偏差，形状的numpy数组 Wc -- Weight matrix of the first "tanh", numpy array of shape (n_a, n_a + n_x) 第一个“tanh”的权重矩阵，形状的numpy数组 bc -- Bias of the first "tanh", numpy array of shape (n_a, 1) 第一个“tanh”的偏置，形状的numpy数组 Wo -- Weight matrix of the output gate, numpy array of shape (n_a, n_a + n_x)输出门的权重矩阵，形状的numpy数组 bo -- Bias of the output gate, numpy array of shape (n_a, 1) 输出门的偏置，形状的numpy数组 Wy -- Weight matrix relating the hidden-state to the output, numpy array of shape (n_y, n_a) 将隐藏状态与输出关联的权重矩阵，形状的numpy数组 by -- Bias relating the hidden-state to the output, numpy array of shape (n_y, 1) 将隐藏状态与输出关联的偏差，numpy形状数组 Returns: a_next -- next hidden state, of shape (n_a, m)下一个隐藏状态 c_next -- next memory state, of shape (n_a, m)下一个记忆状态 yt_pred -- prediction at timestep "t", numpy array of shape (n_y, m)时间步“t”的预测 cache -- tuple of values needed for the backward pass, contains (a_next, c_next, a_prev, c_prev, xt, parameters)向后传递所需的值的元组 Note: ft/it/ot stand for the forget/update/output gates, cct stands for the candidate value (c tilde),cct 代表候选值 c stands for the memory value c代表记忆值 """ # Retrieve parameters from "parameters" 从“参数”中检索参数 Wf = self.parameters["Wf"] bf = self.parameters["bf"] Wi = self.parameters["Wi"] bi = self.parameters["bi"] Wc = self.parameters["Wc"] bc = self.parameters["bc"] Wo = self.parameters["Wo"] bo = self.parameters["bo"] Wy = self.parameters["Wy"] by = self.parameters["by"] # Retrieve dimensions from shapes of xt and Wy 从xt和Wy形状检索尺寸 n_x, m = xt.shape # 时间步 “t”处输入的数据 27 1 n_y, n_a = Wy.shape # 27 50 # Concatenate a_prev and xt (≈3 lines) 连接a_prev和xt concat = np.zeros((n_a + n_x, m)) # (77,1) concat[: n_a, :] = a_prev concat[n_a:, :] = xt # Compute values for ft, it, cct, c_next, ot, a_next using the formulas given figure (4) (≈6 lines) # 使用图（4）中给出的公式计算ft、it、cct、c_next、ot、a_next 的值 ft = sigmoid(np.dot(Wf, concat) + bf) it = sigmoid(np.dot(Wi, concat) + bi) cct = np.tanh(np.dot(Wc, concat) + bc) c_next = np.multiply(ft, c_prev) + np.multiply(it, cct) # ct ot = sigmoid(np.dot(Wo, concat) + bo) a_next = np.multiply(ot, np.tanh(c_next)) # ht # Compute prediction of the LSTM cell (≈1 line)LSTM单元的计算预测 yt_pred = softmax(np.dot(Wy, a_next) + by) # store values needed for backward propagation in cache 在缓存中存储反向传播所需的值 cache = (a_next, c_next, a_prev, c_prev, ft, it, cct, ot, xt) return a_next, c_next, yt_pred, cache def lstm_forward(self, x, a0): """ Implement the forward propagation of the recurrent neural network using an LSTM-cell described in Figure (3). 使用图（3）所示的LSTM单元实现递归神经网络的正向传播 Arguments: x -- Input data for every time-step, of shape (n_x, m, T_x). 输入每个时间步的数据(27,1,12) T_x:单词的长度+1 a0 -- Initial hidden state, of shape (n_a, m) 初始隐藏状态(50,1) parameters -- python dictionary containing: python字典包含： Wf -- Weight matrix of the forget gate, numpy array of shape (n_a, n_a + n_x) bf -- Bias of the forget gate, numpy array of shape (n_a, 1) Wi -- Weight matrix of the update gate, numpy array of shape (n_a, n_a + n_x) bi -- Bias of the update gate, numpy array of shape (n_a, 1) Wc -- Weight matrix of the first "tanh", numpy array of shape