def run(config, args):
args_dict = vars(args)
for parameter, value in config['params'].items():
if parameter not in ('train', 'dev', 'test', 'cuda', 'results_path'):
args_dict[parameter] = value
print(args)
run_experiment(args, config['seed'])
def experiment(self,
experiment_id,
interval_secs,
model_params,
data_params):
"""
Example usage:
python3.4 run.py experiment \
--experiment-id=60_mins_reduce_hash \
--interval-secs=3600 \
--model-params='{"learning_rate":0.05, "l1_regularization_strength":0.00, "l2_regularization_strength":0.5}' \
--data-params='{"shuffle":True, "num_epochs":1, "num_threads":2, batch_size:256}'
:param experiment_id:
:param interval_secs:
:param model_params:
:param data_params:
:return:
"""
run_experiment(
experiment_id=experiment_id,
interval_secs=interval_secs,
model_hyperparams=model_params,
data_hyperparams=data_params)
def run(config, args, n_iter=200):
for i in range(n_iter):
args_dict = vars(args)
args_dict['categorical_features'] = cat_features
args_dict['continuous_features'] = cont_features
args_dict.update(sample_config(config))
print(args)
seed = make_seed()
run_experiment(args, seed)
# --------------------------------
# Load data for word models
# --------------------------------
train, val, test, full, vocab_size, index_word = load_and_shape_data(
'data/Ellen_tweetdata.txt', char_model=False, seq_len=8)
run_experiment(train[0],
train[1],
val[0],
val[1],
test[0],
test[1],
full,
exp_name='validation-ellen',
epochs=20,
char_model=False,
seq_len=8,
depth=1,
dropout=0,
batch_norm=False,
vocab_size=vocab_size,
vocab_dict=index_word)
# --------------------------------
# Load data for word models
# --------------------------------
train, val, test, full, vocab_size, index_word = load_and_shape_data(
'data/Michelleobama_tweetdata.txt', char_model=False, seq_len=8)
train, val, test, full, vocab_size, index_word = load_and_shape_data(
'data/Trump_tweetdata.txt', char_model=True, seq_len=12)
# --------------------------------
# Depth Experiments
# --------------------------------
run_experiment(train[0],
train[1],
val[0],
val[1],
test[0],
test[1],
full,
exp_name='char-depth-1',
epochs=20,
char_model=True,
seq_len=12,
depth=1,
dropout=0,
batch_norm=False,
vocab_size=vocab_size,
vocab_dict=index_word)
run_experiment(train[0],
train[1],
val[0],
val[1],
test[0],
test[1],
full,
conf_name = f"{original_conf_name}_{conf_up_counter}"
else:
print(
f"No compatibles to {original_conf_name} benchmarked, adding new condition {conf_name}."
)
benchmark_dict["results"].update({conf_name: {"n": 0}})
benchmark_dict["results"][conf_name].update({"config": config})
for i in range(args.repetitions):
print(
f"\nRepetition {i + 1}/{args.repetitions} in environment {args.env} with config {conf_name}."
)
reward_history = np.array(
run_experiment(args.env,
config,
init_ray=should_init,
verbose=False).cycle_reward_history)
should_init = False
current_n = benchmark_dict["results"][conf_name]["n"]
if current_n == 0:
means, var = reward_history, np.zeros_like(reward_history)
mean_max, var_max = np.max(reward_history), np.array(0)
else:
means, var = increment_mean_var(
np.array(benchmark_dict["results"][conf_name]["means"]),
np.array(benchmark_dict["results"][conf_name]["var"]),
reward_history, np.zeros_like(reward_history), current_n)
mean_max, var_max = increment_mean_var(
benchmark_dict["results"][conf_name]["mean_max"],
benchmark_dict["results"][conf_name]["var_max"],
# update settings
args.n_agents = n_agents
args.varying_start_value = True
args.timesteps = timesteps
args.compute_frequency_increment_values = False
args.model = model
args.batch_size = batch_size
args.learning_rate = learning_rate
args.normalize_samples = normalize_samples
dict_args = vars(args)
dict_args.update(CONFIG)
args.test_samples = test_samples
run_id, nn_results, nn_scores = run_experiment(args)
plot_parameter_sweep(fit_results, nn_results).savefig(
f"../results/{run_id}_params_sweep.png", dpi=300)
plot_fp_scores(fit_results, nn_results).savefig(
f"../results/{run_id}_fp_scores.png", dpi=300)
dict_args.update(nn_scores)
dict_args.update(fit_scores)
dict_args["runid"] = run_id
df = pd.DataFrame([{
k: v
for k, v in dict_args.items()
if k != "test_samples"
}])