Implemented TestOpenML

Author: carefree0910

_Dist/NeuralNetworks/_Tests/DistNN/TestOpenML.py

@@ -0,0 +1,83 @@
+import os
+import sys
+root_path = os.path.abspath("./")
+if root_path not in sys.path:
+    sys.path.append(root_path)
+
+import copy
+import numpy as np
+import tensorflow as tf
+
+from openml import datasets
+
+from _Dist.NeuralNetworks.g_DistNN.NN import DistAdvanced
+
+GPU_ID = None
+K_RANDOM = 9
+IDS = [
+    38, 46, 179,
+    184, 389, 554,
+    772, 917, 1049,
+    1111, 1120, 1128,
+    293,
+]
+
+
+def swap(arr, i1, i2):
+    arr[..., i1], arr[..., i2] = arr[..., i2], arr[..., i1].copy()
+
+
+def download_data():
+    data_folder = "_Data"
+    idx_folder = os.path.join(data_folder, "idx")
+    if not os.path.isdir(data_folder):
+        os.makedirs(data_folder)
+    if not os.path.isdir(idx_folder):
+        os.makedirs(idx_folder)
+    for idx in IDS:
+        print("Downloading {}".format(idx))
+        data_file = os.path.join(data_folder, "{}.txt".format(idx))
+        idx_file = os.path.join(idx_folder, "{}.npy".format(idx))
+        if os.path.isfile(data_file) and os.path.isfile(idx_file):
+            continue
+        dataset = datasets.get_dataset(idx)
+        data, categorical_idx, names = dataset.get_data(
+            return_categorical_indicator=True,
+            return_attribute_names=True
+        )
+        data = data.toarray() if not isinstance(data, np.ndarray) else data
+        target_idx = names.index(dataset.default_target_attribute)
+        numerical_idx = ~np.array(categorical_idx)
+        swap(numerical_idx, target_idx, -1)
+        swap(data, target_idx, -1)
+        with open(data_file, "w") as file:
+            file.write("\n".join([" ".join(map(lambda n: str(n), line)) for line in data]))
+        np.save(idx_file, numerical_idx)
+
+
+def main():
+    base_params = {
+        "data_info": {},
+        "model_param_settings": {},
+        # "model_param_settings": {"n_epoch": 1, "max_epoch": 1},
+        # "model_structure_settings": {"use_wide_network": False, "use_pruner": False}
+    }
+    config = tf.ConfigProto()
+    config.gpu_options.allow_growth = True
+    if GPU_ID is not None:
+        os.environ["CUDA_VISIBLE_DEVICES"] = GPU_ID
+    base_params["model_param_settings"]["sess_config"] = config
+    for idx in IDS:
+        numerical_idx = np.load("_Data/idx/{}.npy".format(idx))
+        local_params = copy.deepcopy(base_params)
+        local_params["name"] = str(idx)
+        local_params["data_info"]["numerical_idx"] = numerical_idx
+        DistAdvanced(**local_params).empirical_search(cv_rate=0.1, test_rate=0.1).k_random(
+            K_RANDOM, cv_rate=0.1, test_rate=0.1)
+        # DistAdvanced(**local_params).k_random(K_RANDOM, cv_rate=0.1, test_rate=0.1)
+        # DistAdvanced(**local_params).fit()
+
+
+if __name__ == '__main__':
+    download_data()
+    main()