def tune_model_weights():
parser = generate.get_parser_with_args()
parser = add_tune_args(parser)
args = options.parse_args_and_arch(parser)
print(args.model_weights)
n_models = len(args.path.split(CHECKPOINT_PATHS_DELIMITER))
print(n_models)
weight_grid = np.linspace(
args.weight_lower_bound, args.weight_upper_bound, args.n_grid + 1
)
weight_vec_aux = list(itertools.product(weight_grid, weight_grid))
weight_vec = []
for w1, w2 in weight_vec_aux:
weight_sum = w1 + w2
if weight_sum <= 1:
w3 = 1 - weight_sum
weight_vec.append(str(w1) + "," + str(w2) + "," + str(w3))
print(len(weight_vec))
output = pd.DataFrame()
for weight in weight_vec:
args.model_weights = weight
print(args.model_weights)
generate.validate_args(args)
score = generate.generate(args)
print(score)
output = output.append(
{"weight": args.model_weights, "bleu_score": score}, ignore_index=True
)
output.to_csv(args.output_file_name)
return output
def generate_main(data_dir, extra_flags=None):
parser = generate.get_parser_with_args()
args = options.parse_args_and_arch(
parser,
[
"--source-vocab-file",
os.path.join(data_dir, "dictionary-in.txt"),
"--target-vocab-file",
os.path.join(data_dir, "dictionary-out.txt"),
"--source-text-file",
os.path.join(data_dir, "test.in"),
"--target-text-file",
os.path.join(data_dir, "test.out"),
"--path",
os.path.join(data_dir, "checkpoint_last.pt"),
"--beam",
"3",
"--length-penalty",
"0.0",
"--batch-size",
"64",
"--max-len-b",
"5",
"--no-progress-bar",
] + (extra_flags or []),
)
generate.validate_args(args)
generate.generate(args)
def tune_model_weights():
parser = generate.get_parser_with_args()
parser = add_tune_args(parser)
args = options.parse_args_and_arch(parser)
n_models = len(args.path.split(CHECKPOINT_PATHS_DELIMITER))
print(n_models)
print(args.weight_lower_bound)
print(args.weight_upper_bound)
print(args.output_json_best_parameters)
print(args.output_json_best_value)
print(args.num_trails_ax_opt)
def evaluation_function(parameterization):
w1 = parameterization.get("w1")
w2 = parameterization.get("w2")
w3 = parameterization.get("w3")
weight = str(w1) + "," + str(w2) + "," + str(w3)
args.model_weights = weight
generate.validate_args(args)
score = generate.generate(args)
return {"bleu_score": (score, 0.0)}
lower_bound = args.weight_lower_bound
upper_bound = args.weight_upper_bound
best_parameters, values, experiment, model = optimize(
parameters=[
{
"name": "w1",
"type": "range",
"bounds": [lower_bound, upper_bound],
"value_type": "float",
},
{"name": "w2", "type": "range", "bounds": [lower_bound, upper_bound]},
{"name": "w3", "type": "range", "bounds": [lower_bound, upper_bound]},
],
experiment_name="tune_model_weights",
objective_name="bleu_score",
evaluation_function=evaluation_function,
minimize=True, # Optional, defaults to False.
parameter_constraints=[
"w1 + w2 + w3 <= 1",
"w1 + w2 + w3 >= 0.99",
], # Optional.
total_trials=args.num_trails_ax_opt, # Optional.
)
json_file = json.dumps(best_parameters)
with open(args.output_json_best_parameters, "w") as f:
f.write(json_file)
f.close()
json_file = json.dumps(values)
with open(args.output_json_best_value, "w") as f:
f.write(json_file)
f.close()
return best_parameters, values
