att_to_h3 = GRUFork(n_chars, n_hid, random_state)
h1_to_h2 = GRUFork(n_hid, n_hid, random_state)
h1_to_h3 = GRUFork(n_hid, n_hid, random_state)
h2_to_h3 = GRUFork(n_hid, n_hid, random_state)
params += inp_to_h1.get_params()
params += inp_to_h2.get_params()
params += inp_to_h3.get_params()
params += att_to_h1.get_params()
params += att_to_h2.get_params()
params += att_to_h3.get_params()
params += h1_to_h2.get_params()
params += h1_to_h3.get_params()
params += h2_to_h3.get_params()
biases += inp_to_h1.get_biases()
biases += inp_to_h2.get_biases()
biases += inp_to_h3.get_biases()
biases += att_to_h1.get_biases()
biases += att_to_h2.get_biases()
biases += att_to_h3.get_biases()
biases += h1_to_h2.get_biases()
biases += h1_to_h3.get_biases()
biases += h2_to_h3.get_biases()
# 3 to include groundtruth, pixel RNN style
outs_to_v_h1 = GRUFork(3, n_v_proj, random_state)
params += outs_to_v_h1.get_params()
biases += outs_to_v_h1.get_biases()
v_cell1 = GRU(n_v_proj, n_v_proj, random_state)
att_to_h3 = GRUFork(n_chars, n_hid, random_state)
h1_to_h2 = GRUFork(n_hid, n_hid, random_state)
h1_to_h3 = GRUFork(n_hid, n_hid, random_state)
h2_to_h3 = GRUFork(n_hid, n_hid, random_state)
params += inp_to_h1.get_params()
params += inp_to_h2.get_params()
params += inp_to_h3.get_params()
params += att_to_h1.get_params()
params += att_to_h2.get_params()
params += att_to_h3.get_params()
params += h1_to_h2.get_params()
params += h1_to_h3.get_params()
params += h2_to_h3.get_params()
biases += inp_to_h1.get_biases()
biases += inp_to_h2.get_biases()
biases += inp_to_h3.get_biases()
biases += att_to_h1.get_biases()
biases += att_to_h2.get_biases()
biases += att_to_h3.get_biases()
biases += h1_to_h2.get_biases()
biases += h1_to_h3.get_biases()
biases += h2_to_h3.get_biases()
# 3 to include groundtruth, pixel RNN style
outs_to_v_h1 = GRUFork(3, n_v_proj, random_state)
params += outs_to_v_h1.get_params()
biases += outs_to_v_h1.get_biases()
v_cell1 = GRU(n_v_proj, n_v_proj, random_state)
inp_to_h1 = GRUFork(input_dim, n_hid, random_state)
inp_to_h2 = GRUFork(input_dim, n_hid, random_state)
inp_to_h3 = GRUFork(input_dim, n_hid, random_state)
h1_to_h2 = GRUFork(n_hid, n_hid, random_state)
h1_to_h3 = GRUFork(n_hid, n_hid, random_state)
h2_to_h3 = GRUFork(n_hid, n_hid, random_state)
params += inp_to_h1.get_params()
params += inp_to_h2.get_params()
params += inp_to_h3.get_params()
params += h1_to_h2.get_params()
params += h1_to_h3.get_params()
params += h2_to_h3.get_params()
biases += inp_to_h1.get_biases()
biases += inp_to_h2.get_biases()
biases += inp_to_h3.get_biases()
biases += h1_to_h2.get_biases()
biases += h1_to_h3.get_biases()
biases += h2_to_h3.get_biases()
h1_to_outs, = make_weights(n_hid, [n_proj], random_state)
h2_to_outs, = make_weights(n_hid, [n_proj], random_state)
h3_to_outs, = make_weights(n_hid, [n_proj], random_state)
b_to_outs, = make_biases([n_proj])
params += [h1_to_outs, h2_to_outs, h3_to_outs]
biases += [b_to_outs]
mlp1_w, = make_weights(n_inpt_mlp, [n_hid_mlp], random_state)
