def train(data,beta_0, lam = 0.0, max_iter = 100, eps = 0.01):
    N = 2
    data = data.repartition(N).cache()
    beta_k = beta_0
    k = 0
    while k < max_iter:
        start = time.time()
        grad = gradient(data,beta_k,lam)
        print grad
        gamma = lineSearch(lambda x: F(data,beta_k,lam), beta_k, grad, a=0.2,b=0.6)
        beta = beta_k - gamma*gradient(data,beta_k,lam)
        fun_val = F(data,beta,lam)
        print k
        print " %s seconds "%(time.time() - start)

        print fun_val
        print np.linalg.norm(gradient(data,beta,lam))
        beta_k = beta
        k+=1

    return beta, gradNorm, k