import math, numpy, data_read, prob_grad
numpy.set_printoptions(threshold=numpy.nan)
W, T = data_read.read_model()
data = data_read.read_train()
grad = numpy.zeros((26, 26))
import time
t0 = time.time()
for i in range(len(data)):
numpy.add(prob_grad.log_p_tgrad(T, data[i][0], data[i][1], W),\
grad, out=grad)
t1 = time.time()
print(f"Time: {t1-t0}")
avg = numpy.divide(grad, len(data))
for i in range(26):
for j in range(26):
print(avg[j, i])
import numpy, data_read, prob_grad
from scipy.optimize import check_grad
X_y = data_read.read_train()
W, T = data_read.read_model()
def func(T, *args):
t = T.reshape((26, 26))
dataX = args[0]
dataY = args[1]
W = args[2]
return prob_grad.compute_log_p(dataX, dataY, W, t)
def func_prime(T, *args):
t = T.reshape((26, 26))
dataX = args[0]
dataY = args[1]
W = args[2]
return prob_grad.log_p_tgrad(t, dataX, dataY, W).reshape(26 * 26)
x0 = numpy.random.rand(26 * 26)
print(check_grad(func, func_prime, x0, X_y[0][0], X_y[0][1], W))