def load_dataset(self, filename, load_dir):
(dataset_embeddings, dataset_labels,
dataset_imagepaths) = utilities.load_embeddings(
load_path=load_dir, embed_filename=filename)
print("Total Embeddings {}".format(np.array(dataset_embeddings).shape))
print("Total Labels {}".format(np.array(dataset_labels).shape))
print("Total Images {}".format(np.array(dataset_imagepaths).shape))
return dataset_embeddings, dataset_labels, dataset_imagepaths
cv=cv,
verbose=1,
n_jobs=-1)
# Fit the grid search to the data
grid_search.fit(feat_train, lab_train)
if __name__ == '__main__':
# Create the classifier
gbm = GradientBoostMethod()
# Load features and labels
features, labels, ips = utilities.load_embeddings(
load_path=gbm.args["embed_load_dir"],
embed_filename=gbm.args["embed_filename"])
# Train using default params
ml_classifier = MLClassifier(ml_model=gbm.model, model_name=gbm.name)
ml_classifier.train_classifier(features=features,
labels=labels,
save_model=gbm.args["save_model"],
save_name=gbm.args["model_filename"])
# Find the best params
gbm.find_best_model_random(feat_train=features, lab_train=labels)
gbm.find_best_model_grid(feat_train=features, lab_train=labels)
# Train with best found params
ml_classifier.set_model(ml_model=gbm.model)
df_full = pd.concat([df_features, df_labels], axis=1)
df_full['label'] = df_full['label'].astype(str)
#df_full.drop(df_full[df_full["label"] == "Indian"].index, inplace=True)
# Plot
sns.set_style("dark")
fig.add_subplot(rows, cols, i + 1)
sns.scatterplot(x='PC1',
y='PC2',
hue="label",
data=df_full,
palette=sns.color_palette("hls", colors),
alpha=.7,
legend="full",
x_jitter=20,
y_jitter=0).set_title(title)
plt.show()
if __name__ == '__main__':
dataset_embeddings, dataset_labels, dataset_imagepaths = utilities.load_embeddings(
embed_filename="embeddings_ethnicity.pkl")
plot_dimension_reduction(
plots=["PCA", "TSNE", "ISO", "FICA", "LLE", "MDS"],
features=dataset_embeddings,
labels=dataset_labels,
title="Face_Recognition")