def main(args):
config = get_config(args.model, to_bunch=True)
net = load_model_path(args.model, config)
data, loss = test_model(net, config, n_tests=150)
ys = [[] for i in range(net.args.T)]
ys_stim = [[] for i in range(net.args.T)]
for d in data:
context, idx, trial, x, y, out, loss = d
y_ready, y_set, y_go = trial.rsg
y_prod = np.argmax(out >= 1)
t_y = y_set - y_ready
t_p = y_prod - y_set
t_ym = y_go - y_set
ys[context].append((t_y, t_p))
ys_stim[context].append((t_y, t_ym))
ys = [list(zip(*np.array(y))) for y in ys]
ys_stim = [list(zip(*np.array(y))) for y in ys_stim]
for i in range(net.args.T):
plt.scatter(ys_stim[i][0],
ys_stim[i][1],
marker='o',
c='black',
s=20,
edgecolors=cols[i])
plt.scatter(ys[i][0], ys[i][1], color=cols[i], s=15)
plt.xlabel('desired t_p')
plt.ylabel('produced t_p')
plt.show()
J = model['W_f.weight']
v = J.std()
shp = J.shape
model['W_f.weight'] += torch.normal(0, v * .5, shp)
J = model['W_ro.weight']
v = J.std()
shp = J.shape
model['W_ro.weight'] += torch.normal(0, v * .5, shp)
config = fill_undefined_args(args, config, overwrite_none=True)
net = load_model_path(args.model, config=config)
if args.test_all:
_, loss2 = test_model(net, config)
print('avg summed loss (all):', loss2)
if not args.no_plot:
data, loss = test_model(net, config, n_tests=6)
print('avg summed loss (plotted):', loss)
run_id = '/'.join(args.model.split('/')[-3:-1])
fig, ax = plt.subplots(2, 3, sharex=True, sharey=True, figsize=(12, 7))
if 'goals' in config.dataset:
p_fn = get_potential(config)
for i, ax in enumerate(fig.axes):
ix, x, y, z, loss = data[i]
xr = np.arange(len(x))
intervals = [{}, {}, {}]
for j, dset in enumerate(dsets):
subset = dt[dt.dset == dset]
for iterr in range(len(subset)):
job_id = subset.iloc[iterr].slurm_id
model_folder = os.path.join('..', 'logs', run_id, str(job_id))
model_path = os.path.join(model_folder, 'model_best.pth')
config = get_config(model_path, ctype='model', to_bunch=True)
config.m_noise = 0
config.dataset = dset_map[config.dataset]
net = load_model_path(model_path, config=config)
data, loss = test_model(net, config, n_tests=200, dset_base='../')
dset = load_rb(os.path.join('..', config.dataset))
distr = {}
for k in range(len(data)):
dset_idx, x, _, z, _ = data[k]
r, s, g = dset[dset_idx][2]
t_first = torch.nonzero(z >= 1)
if len(t_first) > 0:
t_first = t_first[0, 0]
else:
t_first = len(x)
val = np.array(t_first - s - 5)
model = torch.load(f)
if args.noise != 0:
J = model['W_f.weight']
v = J.std()
shp = J.shape
model['W_f.weight'] += torch.normal(0, v * .5, shp)
J = model['W_ro.weight']
v = J.std()
shp = J.shape
model['W_ro.weight'] += torch.normal(0, v * .5, shp)
net = load_model_path(args.model, params={'dset': args.dataset, 'out_act': args.out_act})
dset = load_rb(args.dataset)
data = test_model(net, dset, n_tests=0)
run_id = '/'.join(args.model.split('/')[-3:-1])
fig, ax = plt.subplots(3,4,sharex=True, sharey=True, figsize=(12,7))
for i, ax in enumerate(fig.axes):
ix, x, y, z, loss = data[i]
xr = np.arange(len(x))
ax.axvline(x=0, color='dimgray', alpha = 1)
ax.axhline(y=0, color='dimgray', alpha = 1)
ax.grid(True, which='major', lw=1, color='lightgray', alpha=0.4)
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
ax.spines['left'].set_visible(False)
default=None,
help='path to config file if custom')
args = parser.parse_args()
if args.config is None:
config = get_config(args.model, ctype='model')
else:
config = json.load(open(args.config, 'r'))
config = update_args(args, config)
dsets = config.dataset
net = load_model_path(args.model, config=config)
# assuming config is in the same folder as the model
if args.test_all:
_, loss = test_model(net, config)
print('avg summed loss (all):', loss)
if not args.no_plot:
data, t_losses = test_model(net, config, n_tests=12)
print('avg losses:')
for t, j in t_losses.items():
print(t + ': ' + str(j))
run_id = '/'.join(args.model.split('/')[-3:-1])
fig, ax = plt.subplots(3, 4, sharex=False, sharey=False, figsize=(12, 8))
for i, ax in enumerate(fig.axes):
context, ix, trial, x, y, z, loss = data[i]
xr = np.arange(x.shape[-1])
ax.axvline(x=0, color='dimgray', alpha=1)
with open(args.model, 'rb') as f:
m_dict = torch.load(f)
J = m_dict['W_f.weight']
v = J.std()
shp = J.shape
m_dict['W_f.weight'] += torch.normal(0, v * .01, shp)
J = m_dict['W_ro.weight']
v = J.std()
shp = J.shape
m_dict['W_ro.weight'] += torch.normal(0, v * .01, shp)
dset = load_rb(args.dset)
test_data = test_model(m_dict, dset, n_tests=200)
i, xs, ys, zs, losses = list(zip(*test_data))
print(np.mean(losses))
# bunch = Bunch()
# bunch.N = 250
# bunch.D = 250
# bunch.O = 1
# bunch.res_init_type = 'gaussian'
# bunch.res_init_params = {'std': 1.5}
# bunch.reservoir_seed = 0
# net = Network(bunch)
# nums = ['146', '152', '158', '164', '170', '176', '182', '188']