def run_experiment(dataset_id: str):
ROOT_DIR = Path(__file__).parents[2]
dataset_folder = ROOT_DIR / "datasets" / dataset_id
outputs_folder = ROOT_DIR / "outputs" / dataset_id
output_filepath = outputs_folder / "scoring.pickled"
selected_samples_filepath = outputs_folder / "selected.csv"
graphs_index = load_graph_index(dataset_id)
print(f"> Testing predicting the team labels for {dataset_id}...")
# This dataset is balanced and does not need balancing
print(
"Current number of players in each team:\n", graphs_index.label.value_counts()
)
selected_graphs = graphs_index
selected_graphs.graph_file.to_csv(selected_samples_filepath)
print(f"Generating GraKeL graphs for {len(selected_graphs)} files")
selected_graphs = build_grakel_graphs(selected_graphs, dataset_folder)
cv_sets = get_fixed_CV_sets(selected_graphs, selected_graphs.label)
print(f"Generated {len(cv_sets)} cross-validation train/test sets.")
results = test_prediction_on_classifiers(
selected_graphs[POKEMON_GO_DATA_COLUMNS], selected_graphs.label, cv_sets
)
pna_results = test_prediction_on_classifiers(
selected_graphs[NETWORK_METRIC_NAMES],
selected_graphs.label,
cv_sets,
test_prefix="PNA-",
)
pna_results["time"] = selected_graphs.timings_PNA.sum()
results = results.append(pna_results, ignore_index=True)
results = results.append(
test_prediction_on_Grakel_kernels(
selected_graphs, "label", cv_sets, ignore_kernels={"GK-GSamp"}
),
ignore_index=True,
)
results = results.append(
test_prediction_on_kernels(selected_graphs, outputs_folder, "label", cv_sets),
ignore_index=True,
)
print("Saving scoring to:", output_filepath)
results.to_pickle(output_filepath)
def main(output_folder: str, kernel: str, y_column: str, to_level: int):
output_path = Path(output_folder)
dataset_id = output_path.name
graphs_index = load_graph_index(dataset_id)
selected_samples_filepath = output_path / "selected.csv"
selected_graphfiles = pd.read_csv(selected_samples_filepath, index_col=0)
selected_graphs = graphs_index.iloc[selected_graphfiles.index].copy()
print(f"Training {kernel}-{to_level} on {len(selected_graphs)} {dataset_id} graphs...")
clf = train_provenance_kernel_pipeline(
selected_graphs, output_path, kernel, to_level, y_column
)
models_folder = output_path / "models"
models_folder.mkdir(parents=True, exist_ok=True)
model_filepath = models_folder / f"model_{kernel}_{to_level}.pickled"
with model_filepath.open("wb") as f:
pickle.dump(clf, f)
def run_experiment(dataset_id: str):
ROOT_DIR = Path(__file__).parents[2]
dataset_folder = ROOT_DIR / "datasets" / dataset_id
outputs_folder = ROOT_DIR / "outputs" / dataset_id
output_filepath = outputs_folder / "scoring.pickled"
selected_samples_filepath = outputs_folder / "selected.csv"
graphs_index = load_graph_index(dataset_id)
print(f"> Testing predicting the data quality labels for {dataset_id}...")
# Balancing the dataset on the trusted attribute
print("Current number of trusted values:\n",
graphs_index.trusted.value_counts())
selected_graphs = graphs_index[graphs_index.trusted == False]
selected_graphs = selected_graphs.append(
graphs_index[graphs_index.trusted == True].sample(
len(selected_graphs)))
print(
"Number of trusted values in selected graphs:\n",
selected_graphs.trusted.value_counts(),
)
selected_graphs.graph_file.to_csv(selected_samples_filepath)
print(f"Generating GraKeL graphs for {len(selected_graphs)} files")
selected_graphs = build_grakel_graphs(selected_graphs, dataset_folder)
# ---------- Ayah's code ----------------------------------------------
selected_graphs_filepath = outputs_folder / f"{dataset_id}.pickled"
selected_graphs_filepath_2 = outputs_folder / f"{dataset_id}-Analytics.pickled"
index = selected_graphs.columns.get_loc(NETWORK_METRIC_NAMES[0])
applicationdata = selected_graphs.columns[:index]
X = selected_graphs[applicationdata]
y = selected_graphs.trusted
data = pd.concat([X, y], axis=1)
data = data.rename(columns={"trusted": "Class"})
data.to_pickle(selected_graphs_filepath)
X = selected_graphs[NETWORK_METRIC_NAMES]
y = selected_graphs.trusted
data = pd.concat([X, y], axis=1)
data = data.rename(columns={"trusted": "Class"})
data.to_pickle(selected_graphs_filepath_2)
# --------------------------------------------------------------------
cv_sets = get_fixed_CV_sets(selected_graphs, selected_graphs.trusted)
print(f"Generated {len(cv_sets)} cross-validation train/test sets.")
results = test_prediction_on_classifiers(
selected_graphs[NETWORK_METRIC_NAMES],
selected_graphs.trusted,
cv_sets,
test_prefix="PNA-",
)
results["time"] = selected_graphs.timings_PNA.sum()
results = results.append(
test_prediction_on_Grakel_kernels(selected_graphs, "trusted", cv_sets),
ignore_index=True,
)
results = results.append(
test_prediction_on_kernels(selected_graphs, outputs_folder, "trusted",
cv_sets),
ignore_index=True,
)
print("Saving scoring to:", output_filepath)
results.to_pickle(output_filepath)
test_prediction_on_classifiers,
test_prediction_on_kernels,
test_prediction_on_Grakel_kernels,
)
from scripts.utils import load_graph_index
ROOT_DIR = Path(__file__).parents[2]
data_folder = ROOT_DIR / "datasets"
dataset_id = "MIMIC-PXC7"
dataset_folder = ROOT_DIR / "datasets" / dataset_id
outputs_folder = ROOT_DIR / "outputs" / dataset_id
output_filepath = outputs_folder / "scoring.pickled"
selected_samples_filepath = outputs_folder / "selected.csv"
graphs_index = load_graph_index(dataset_id)
print("> Testing predicting a patient is dead at the end of this admission")
if selected_samples_filepath.exists():
# Loading the previously saved balanced dataset to reproduce the same experiment
selected_graphfiles = pd.read_csv(selected_samples_filepath, index_col=0)
selected_graphs = graphs_index.iloc[selected_graphfiles.index].copy()
else:
# This is the first time we run this experiment
# Selecting relevant graphs and balancing the dataset
print(" - Current number of dead values:\n",
graphs_index.dead.value_counts())
selected_graphs = graphs_index[graphs_index.dead == True]
selected_graphs = selected_graphs.append(
graphs_index[graphs_index.dead == False].sample(len(selected_graphs)))
def run_experiment(dataset_id: str):
ROOT_DIR = Path(__file__).parents[2]
dataset_folder = ROOT_DIR / "datasets" / dataset_id
outputs_folder = ROOT_DIR / "outputs" / dataset_id
output_filepath = outputs_folder / "scoring.pickled"
selected_samples_filepath = outputs_folder / "selected.csv"
graphs_index = load_graph_index(dataset_id)
print(f"> Testing predicting the data quality labels for {dataset_id}...")
if selected_samples_filepath.exists():
# Loading the previously saved balanced dataset to reproduce the same experiment
selected_graphfiles = pd.read_csv(selected_samples_filepath,
index_col=0)
selected_graphs = graphs_index.iloc[selected_graphfiles.index].copy()
else:
# This is the first time we run this experiment
# Balancing the dataset on the trusted attribute
print(
" - Current number of trusted values:\n",
graphs_index.trusted.value_counts(),
)
selected_graphs = graphs_index[graphs_index.trusted == False]
selected_graphs = selected_graphs.append(
graphs_index[graphs_index.trusted == True].sample(
len(selected_graphs)))
print(
" - Number of trusted values in selected graphs:\n",
selected_graphs.trusted.value_counts(),
)
# saving the list of selected graphs for later reproduction of this experiment
selected_graphs.graph_file.to_csv(selected_samples_filepath)
print(f"> Generating GraKeL graphs for {len(selected_graphs)} files")
selected_graphs = build_grakel_graphs(selected_graphs, dataset_folder)
cv_sets = get_fixed_CV_sets(selected_graphs,
selected_graphs.trusted,
output_path=outputs_folder)
print(f"> Got {len(cv_sets)} cross-validation train/test sets.")
results = test_prediction_on_classifiers(
selected_graphs[NETWORK_METRIC_NAMES],
outputs_folder,
selected_graphs.trusted,
cv_sets,
test_prefix="PNA-",
)
results["time"] = selected_graphs.timings_PNA.sum()
results = results.append(
test_prediction_on_Grakel_kernels(selected_graphs, outputs_folder,
"trusted", cv_sets),
ignore_index=True,
)
results = results.append(
test_prediction_on_kernels(selected_graphs, outputs_folder, "trusted",
cv_sets),
ignore_index=True,
)
print("> Saving scoring to:", output_filepath)
results.to_pickle(output_filepath)