22425200b7220b2b410c9aa5ebc4c783f6cfe5df,lxmls/deep_learning/rnn.py,LSTM,__init__,#LSTM#,271

Before Change

    def __init__(self, W_e, n_hidd, n_tags):

        // Dimension of the embeddings
        n_emb = W_e.shape[0]

        // MODEL PARAMETERS
        W_x = np.random.uniform(size=(4*n_hidd, n_emb))   // RNN Input layer
        W_h = np.random.uniform(size=(4*n_hidd, n_hidd))  // RNN recurrent var 
        W_c = np.random.uniform(size=(3*n_hidd, n_hidd))  // Second recurrent var 
        W_y = np.random.uniform(size=(n_tags, n_hidd))    // Output layer
        // Cast to theano GPU-compatible type
        W_e = W_e.astype(theano.config.floatX)
        W_x = W_x.astype(theano.config.floatX)
        W_h = W_h.astype(theano.config.floatX)
        W_c = W_c.astype(theano.config.floatX)
        W_y = W_y.astype(theano.config.floatX)
        // Store as shared parameters
        _W_e = theano.shared(W_e, borrow=True)
        _W_x = theano.shared(W_x, borrow=True)
        _W_h = theano.shared(W_h, borrow=True)
        _W_c = theano.shared(W_c, borrow=True)
        _W_y = theano.shared(W_y, borrow=True)

        // Class variables
        self.n_hidd = n_hidd
        self.param  = [_W_e, _W_x, _W_h, _W_c, _W_y]

    def _forward(self, _x, _h0=None, _c0=None):

After Change

            np.random.seed(seed)

        // MODEL PARAMETERS
        W_e = 0.01*np.random.uniform(size=(n_emb, n_words))    // Embedding layer 
        W_x = np.random.uniform(size=(4*n_hidd, n_emb))   // RNN Input layer
        W_h = np.random.uniform(size=(4*n_hidd, n_hidd))  // RNN recurrent var 
        W_c = np.random.uniform(size=(3*n_hidd, n_hidd))  // Second recurrent var 
        W_y = np.random.uniform(size=(n_tags, n_hidd))    // Output layer
        // Cast to theano GPU-compatible type
        W_e = W_e.astype(theano.config.floatX)
        W_x = W_x.astype(theano.config.floatX)
        W_h = W_h.astype(theano.config.floatX)
        W_c = W_c.astype(theano.config.floatX)
        W_y = W_y.astype(theano.config.floatX)
        // Store as shared parameters
        _W_e = theano.shared(W_e, borrow=True)
        _W_x = theano.shared(W_x, borrow=True)
        _W_h = theano.shared(W_h, borrow=True)
        _W_c = theano.shared(W_c, borrow=True)
        _W_y = theano.shared(W_y, borrow=True)

        // Class variables
        self.n_hidd = n_hidd
        self.param  = [_W_e, _W_x, _W_h, _W_c, _W_y]

    def _forward(self, _x, _h0=None, _c0=None):

Instances