n_special = len(special)
n_ctx = context_size_event + context_size_effect
n_vocab = len(text_encoder.encoder) + n_ctx
print(data_loader.__dict__.keys())
opt.net.vSize = n_vocab
print("Building Model")
print(opt.exp)
model = models.make_model(
opt, n_vocab, n_ctx, 0, load=False, return_acts=False, return_probs=True)
models.load_state_dict(model, model_stuff["state_dict"])
if config.gpu_mode:
print("Pushing to GPU: {}".format(config.gpu_index))
cfg.device = config.gpu_index
cfg.do_gpu = True
torch.cuda.set_device(cfg.device)
model.cuda(cfg.device)
print("Done.")
model.eval()
device = cfg.device
model.to(device)
random.seed(args.seed)
print("Original number of evaluation sequences: {}".format(
len(data_loader.sequences[split]["total"])))
adata.prune_data_for_evaluation(
data_loader, ["<{}>".format(cat) for cat in opt.eval.categories],
split)
print("Pruned number of evaluation sequences for subset: {}".format(
len(data_loader.sequences[split]["total"])))
print("Building Model")
model = models.make_model(opt, n_vocab, n_ctx, n_special, load=False)
print("Loading Weights")
models.load_state_dict(model, model_file["state_dict"])
print("Done Loading Weights")
model.eval()
# Initialize variable for # of examples to cycle through
data.set_max_sizes(data_loader, force_split=split)
evaluator = evaluate.make_evaluator(opt, model, data_loader)
evaluator.batch_variables["split"] = split
model.cuda(cfg.device)
loss = evaluator.epoch(opt, model, data_loader, split)
data.save_eval_file(opt, loss, "losses", split=split)