def call_plot_hyp(data_set, params):
# Load range of hyper parameters tested on to plot
try:
model_hyp_range = json.load(
open('experiments/config/%s_hypRange.conf' % data_set, 'r')
)
except IOError:
model_hyp_range = json.load(
open('experiments/config/default_hypRange.conf', 'r')
)
for meth in params["methods"]:
meth_hyp_range = model_hyp_range[meth]
exp_param = params["experiments"]
plot_util.plot_hyp(meth_hyp_range.keys(), exp_param,
meth, data_set,
s_sch=params["samp_scheme"])
def call_plot_hyp(data_set, params):
"""Function to plot the result of hyperparameter search
Attributes:
data_set (str): Name of the dataset to be used for the experiment
params (dict): Dictionary of parameters necessary for running the experiment
"""
# Load range of hyper parameters tested on to plot
try:
model_hyp_range = json.load(
open('experiments/config/%s_hypRange.conf' % data_set, 'r'))
except IOError:
model_hyp_range = json.load(
open('experiments/config/default_hypRange.conf', 'r'))
for meth in params["methods"]:
meth_hyp_range = model_hyp_range[meth]
exp_param = params["experiments"]
plot_util.plot_hyp(meth_hyp_range.keys(),
exp_param,
meth,
data_set,
s_sch=params["samp_scheme"])
def choose_best_hyp(data_set, graphs, params):
# Load range of hyper parameters to test on
try:
model_hyp_range = json.load(
open('experiments/config/%s_hypRange2.conf' % data_set, 'r')
)
except IOError:
model_hyp_range = json.load(
open('experiments/config/default_hypRange.conf', 'r')
)
# Test each hyperparameter for each method and store the best
for meth in params["methods"]:
dim = 128
MethClass = getattr(
importlib.import_module("embedding.%s" % meth),
methClassMap[meth]
)
meth_hyp_range = model_hyp_range[meth]
gr_max, lp_max = 0, 0
gr_hyp, lp_hyp = 0, 0
gr_hyp, lp_hyp = {meth: {}}, {meth: {}}
# Test each hyperparameter
ev_cols = ["GR MAP", "LP MAP"]
hyp_df = pd.DataFrame(
columns=list(meth_hyp_range.keys()) + ev_cols + ["Round Id"]
)
hyp_r_idx = 0
print()
#print(meth_hyp_range.values())
print()
for hyp in list(itertools.product(*meth_hyp_range.values())):
#hyp_d = {"d": dim}
hyp_d = {}
hyp_d.update(dict(zip(meth_hyp_range.keys(), hyp)))
print(hyp_d)
if meth == "dynAE" or meth == "dynRNN" or meth == "dynAERNN":
hyp_d.update({
"modelfile": [
"./intermediate/encoder_model_%s_%d.json" % (data_set, dim),
"./intermediate/decoder_model_%s_%d.json" % (data_set, dim)
],
"weightfile": [
"./intermediate/encoder_weights_%s_%d.hdf5" % (data_set, dim),
"./intermediate/decoder_weights_%s_%d.hdf5" % (data_set, dim)
]
})
# elif meth == "gf" or meth == "node2vec":
# hyp_d.update({"data_set": data_set})
MethObj = MethClass(dim, hyp_d)
gr, lp = run_exps(MethObj, meth, dim, graphs,
data_set, params)
gr_m, lp_m = np.mean(gr), np.mean(lp)
gr_max, gr_hyp[meth] = get_max(gr_m, gr_max, hyp_d, gr_hyp[meth])
lp_max, lp_hyp[meth] = get_max(lp_m, lp_max, hyp_d, lp_hyp[meth])
hyp_df_row = dict(zip(meth_hyp_range.keys(), hyp))
for r_id in range(params["rounds"]):
hyp_df.loc[hyp_r_idx, meth_hyp_range.keys()] = \
pd.Series(hyp_df_row)
# hyp_df.loc[hyp_r_idx, ev_cols + ["Round Id"]] = \
# [gr[min(r_id, len(gr) -1)], lp[r_id], r_id]
hyp_df.loc[hyp_r_idx, ev_cols + ["Round Id"]] = \
[np.mean(np.array(gr)), np.mean(np.array(lp)), r_id]
hyp_r_idx += 1
exp_param = params["experiments"]
for exp in exp_param:
hyp_df.to_hdf(
"intermediate/%s_%s_%s_%s_hyp.h5" % (data_set, meth,
exp,
params["samp_scheme"]),
"df"
)
plot_util.plot_hyp(meth_hyp_range.keys(), exp_param,
meth, data_set, s_sch=params["samp_scheme"])
# Store the best hyperparameter
opt_hyp_f_pre = 'experiments/config/%s_%s_%s' % (
data_set,
meth,
params["samp_scheme"]
)
if gr_max:
with open('%s_gr.conf' % opt_hyp_f_pre, 'w') as f:
f.write(json.dumps(gr_hyp, indent=4))
if lp_max:
with open('%s_lp.conf' % opt_hyp_f_pre, 'w') as f:
f.write(json.dumps(lp_hyp, indent=4))