feature_output_file = \
os.path.join(feature_output_path,
"{}_all_featurized_data.csv".format(data_name))
if os.path.exists(feature_output_file):
df = pd.read_csv(feature_output_file, index_col=0)
else:
df_init = load_glass_ternary_landolt()
prof = Profile()
prof.enable()
featzer = Featurize()
df_feats = featzer.featurize_formula(df_init, featurizers="all")
prep = PreProcess(max_colnull=0.1)
df = prep.preprocess(df_feats)
prof.create_stats()
print("featurize time:\n")
pstats.Stats(prof).strip_dirs().sort_stats("time").print_stats(5)
if os.path.exists(feature_output_path):
print("output path: {} exists!".format(feature_output_path))
else:
os.makedirs(feature_output_path)
print("create output path: {} successful!".format(feature_output_path))
prof.dump_stats(
os.path.join(feature_output_path,
"cProfile_for_featurize_{}.log".format(data_name)))
# actual pipeline:
df_init = load_castelli_perovskites()
if LIMIT and LIMIT < len(df_init):
df_init = df_init.iloc[np.random.choice(len(df_init), LIMIT,
replace=False)]
featzer = Featurize(ignore_cols=IGNORE_THESE_COLUMNS,
exclude=EXCLUDED_FEATURIZERS,
multiindex=MULTIINDEX,
drop_featurized_col=True)
df = featzer.auto_featurize(df_init,
input_cols=FEATUREIZE_THESE_COLUMNS,
guess_oxidstates=True)
prep = PreProcess(target=TARGET)
df = prep.preprocess(df)
X_train, X_test, y_train, y_test = train_test_split(df.drop(TARGET, axis=1),
df[TARGET])
print('start timing...')
start_time = time()
tpot = TPOTAutoML(mode=MODE,
max_time_mins=TIMEOUT_MINS,
generations=GENERATIONS,
population_size=POPULATION_SIZE,
scoring=SCORING,
random_state=RS,
feature_names=df.drop(TARGET, axis=1).columns,
n_jobs=1,
target = 'gap expt'
RS = 24
mode = 'regression'
MULTIINDEX = True
if MULTIINDEX:
target = ('Input Data', target)
df_init = load_expt_gap()
featzer = Featurize(exclude=['CohesiveEnergy', 'AtomicPackingEfficiency'],
multiindex=MULTIINDEX)
df = featzer.featurize_formula(df_init,
featurizers='all',
guess_oxidstates=False)
prep = PreProcess(target=target)
df = prep.preprocess(df)
print(df.head())
df.to_csv('test.csv')
X_train, X_test, y_train, y_test = train_test_split(df.drop(target, axis=1),
df[target])
model = RandomForestRegressor(n_estimators=100,
bootstrap=False,
max_features=0.8,
min_samples_leaf=1,
min_samples_split=4,
random_state=RS)