def test(top_models = 3):
horizontal = [int(x.split('.')[0]) for x in df.columns.values]
y = np.round(100*model.predict_proba(df), 2)
#pred = encoder.inverse_transform(np.argmax(y , axis = 1))
idx = y.argsort()[:,::-1][:,:top_models]
label_path = os.path.join(config._get_path('graphics'), 'labeling')
min_array = []
mean_array = []
max_array = []
for cl in range(len(encoder.classes_)):
min_array.append(np.min(all_y[cl]))
mean_array.append(np.mean(all_y[cl]))
max_array.append(np.max(all_y[cl]))
if not os.path.exists(label_path):
os.makedirs(label_path)
#https://github.com/matplotlib/matplotlib/issues/8519#issuecomment-608434198
plt.ioff()
for i in range(idx.shape[0]):
sample = df.iloc[i]
k = 1
plt.figure(figsize=(top_models*5, 7))
for j in idx[i]:
m = encoder.inverse_transform([j])[0]
plt.subplot(1, top_models, k)
plt.title("{}, Probability: {}%".format(m, y[i][j]))
plt.xlabel("Wavelenght (1/cm)")
plt.plot(horizontal, sample.values, '-',
color= purple, label= "Sample")
plt.plot(horizontal, mean_array[j], '-', color= blue, label= m)
plt.fill_between(horizontal, min_array[j], max_array[j],
alpha= 0.25,color=blue)
plt.legend(loc="best")
k += 1
sample = None
plt.savefig(os.path.join(label_path, 'prediction_{}.png'.format(i)),
dpi = 300,
bbox_inches = "tight")
plt.close()
return None
#Sklearn Model Selection
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import StratifiedKFold
#Project modules
from utils import file_name as f_name
from utils import append_time
from param_grid import neural_grid, classical_grid
from pipeline import build_pipe
from baseline import als
#Config module
import config
seed = config._seed()
gs_folder = config._get_path('grid_search')
def search(scaler='',
baseline=True,
pca=True,
over_sample=True,
param_grid=classical_grid(),
prefix='',
n_jobs=1,
save=True):
print('Loading training set...')
X_train = pd.read_csv(os.path.join('data', 'X_train.csv'))
y_train = pd.read_csv(os.path.join('data', 'y_train.csv')).values.ravel()
def best_results():
estimator_path = {}
models = {}
for scaler, baseline, pca, over, nn in product([False, True], repeat=5):
file_name = f_name(nn=nn,
baseline=baseline,
scaler=scaler,
pca=pca,
over_sample=over)
file_path = os.path.join(config._get_path('grid_search'), file_name)
if os.path.isfile(file_path):
df = pd.read_csv(file_path)
replace = True
if nn:
row = df.iloc[0]
if 'NN' in models:
if models['NN'][4] <= -row["neg_log_loss"]:
replace = False
if replace:
models['NN'] = [
int(baseline),
int(scaler),
int(pca),
int(over), -row["neg_log_loss"], row["std"]
]
estimator_path['NN'] = file_path
else:
for key in classical_models:
row = next(r for _, r in df.iterrows()
if key in r["estimator"])
replace = True
if classical_models[key] in models:
if (models[classical_models[key]][4] <=
-row["neg_log_loss"]):
replace = False
if replace:
models[classical_models[key]] = [
int(baseline),
int(scaler),
int(pca),
int(over), -row["neg_log_loss"], row["std"]
]
estimator_path[key] = file_path
else:
print(file_name + " does not exists, please run the gridSearch!")
break
data = []
idxs = []
for key in models:
data.append(models[key])
idxs.append(key)
df = pd.DataFrame(data=data,
columns=[
"Baseline", "Standard scaler", "PCA (99%)",
"Over sample", "Log-loss", "Standard Deviation"
],
index=idxs)
df.sort_values(by=["Log-loss"], inplace=True)
return df, estimator_path
#load project modules
import config
from param_grid import build_nn
from utils import append_time, build_row
from baseline import als
from tensorflow.keras.wrappers.scikit_learn import KerasClassifier
from tensorflow.keras.callbacks import EarlyStopping
import warnings
warnings.filterwarnings("ignore", category=DeprecationWarning)
#Beta version!
mccv_path = config._get_path('mccv')
def results_total(X, name, sufix, temp=True): #Arrumar!
posfix = '_temp' if temp else ''
filepath = os.path.join(mccv_path, name, sufix + posfix + '.csv')
pd.DataFrame(data=X,
columns=[
'Cross_Entropy_train', 'Cross_Entropy_val',
'Accuracy_train', 'Accuracy_val'
]).to_csv(filepath, index=False)
from matplotlib import pyplot as plt
from matplotlib.colors import ListedColormap
import seaborn as sns
#Project packages
import config
from utils import classes_names
from table import best_results
#Still beta, several updates required!
#Best model path:
best_path = os.path.join('results', 'mccv', 'baseline_over_SVC_linear_100.0',
'detailed_score.csv')
mccv_path = config._get_path('mccv')
graphics_path = config._get_path('graphics')
def gs_heatmap(output_name='gs_table'):
df, _ = best_results()
c_map = plt.get_cmap('YlGnBu')
c_map = ListedColormap(c_map(np.linspace(0.1, 0.7, 256)))
fig, ax = plt.subplots(figsize=(12, 7))
heat = sns.heatmap(df,
annot=True,
linewidths=1,
cmap=c_map,