plan to implement rnn from scratch

from_scratch_rnn.py

# class RNN:
#   def __init__(self, input_shape, hidden_shape, output_shape):
#     self.input_shape = input_shape
#     self.hidden_shape = hidden_shape
#     self.output_shape = output_shape
#     self.initialise()
#     self.bptt_truncate = 2

#   def initialise(self):
#     self.Whx = np.random.normal(0, 0.01, shape=(self.hidden_shape, self.input_shape))
#     self.Whh = np.random.normal(0, 0.01, shape=(self.hidden_shape, self.hidden_shape))
#     self.bh = np.random.normal(0, 0.01, shape=(self.hidden_shape))
#     self.Wqh = np.random.normal(0, 0.01, shape=(self.output_shape, self.hidden_shape))
#     self.bq = np.random.normal(0, 0.01, shape=(self.output_shape))

#   def tanh(Z):
#     return (np.exp(Z)-np.exp(-Z))/(np.exp(Z)-np.exp(-Z))

#   def softmax(Z):
#     e_x = np.exp(Z - np.max(Z))
#       return e_x / e_x.sum(axis=0)
  
#   def forward(self, x):
#     prev_memory = np.zeros(shape=(self.hidden_shape,1))
#     cache = {}
#     for t in range(x.shape[0]):
#       WX = np.dot(self.Whx, x[i])
#       UH = np.dot(self.Whh, prev_memory)
#       H = WX + UH + self.bh
#       H_act = self.tanh(H)
#       Q = np.dot(self.Wqh, H_act) + self.bq
#       Q_act = self.softmax(Q)
#       prev_memory = H_act
#       cache["WX" + str(t)] = WX
#       cache["UH" + str(t)] = UH
#       cache["H" + str(t)] = H
#       cache["H_act" + str(t)] = H_act
#       cache["Q" + str(t)] = Q
#       cache["Q_act" + str(t)] = Q_act
#     return cache

#   def loss(ypred, ytrue):


#   def backward():

#   def train(self, TrainX, TrainY, ValX, ValY, epochs=1, lr=0.001):
#     assert(epochs>=0)
#     for e in range(epochs):
#       #for training
#       Tcache = self.forward(TrainX)
#       weights = self.backward(Tcache, TrainX, TrainY)
#       self.update_weights(weights)

#       #for metrics
#       Tcache = self.forward(TrainX)
#       TL, TA = self.metrics(Tcache, TrainY)
#       Vcache = self.forward(ValX)
#       VL, VA = self.metrics(Vcache, ValY)
#       print("Epoch: %d | Train loss: %.4f | Train acc: %.4f | Val loss: %.4f | Val acc: %.4f" % (e, TL, TA, VL, VA))