libmultilabel 0.5.1__tar.gz → 0.5.2__tar.gz

{libmultilabel-0.5.1 → libmultilabel-0.5.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: libmultilabel
-Version: 0.5.1
+Version: 0.5.2
 Summary: A library for multi-label text classification
 Home-page: https://github.com/ASUS-AICS/LibMultiLabel
 Author: LibMultiLabel Team

{libmultilabel-0.5.1 → libmultilabel-0.5.2}/libmultilabel/common_utils.py

@@ -3,6 +3,7 @@ import json
 import logging
 import os
 import time
+from functools import wraps
 
 import numpy as np
 
@@ -41,36 +42,6 @@ class AttributeDict(dict):
         return {k: self[k] for k in self._used}
 
 
-class Timer(object):
-    """Computes elasped time."""
-
-    def __init__(self):
-        self.reset()
-
-    def reset(self):
-        self.running = True
-        self.total = 0
-        self.start = time.time()
-        return self
-
-    def resume(self):
-        if not self.running:
-            self.running = True
-            self.start = time.time()
-        return self
-
-    def stop(self):
-        if self.running:
-            self.running = False
-            self.total += time.time() - self.start
-        return self
-
-    def time(self):
-        if self.running:
-            return self.total + time.time() - self.start
-        return self.total
-
-
 def dump_log(log_path, metrics=None, split=None, config=None):
     """Write log including the used items of config and the evaluation scores.
 
@@ -156,3 +127,17 @@ def is_multiclass_dataset(dataset, label="label"):
             a multi-class problem."""
         )
     return ratio == 1.0
+
+
+def timer(func):
+    """Log info-level wall time"""
+
+    @wraps(func)
+    def wrapper(*args, **kwargs):
+        start_time = time.time()
+        value = func(*args, **kwargs)
+        wall_time = time.time() - start_time
+        logging.info(f"{repr(func.__name__)} finished in {wall_time:.2f} seconds")
+        return value
+
+    return wrapper

{libmultilabel-0.5.1 → libmultilabel-0.5.2}/libmultilabel/nn/data_utils.py

@@ -2,7 +2,6 @@ import csv
 import gc
 import logging
 import warnings
-from concurrent.futures import ProcessPoolExecutor
 
 import pandas as pd
 import torch
@@ -159,7 +158,7 @@ def _load_raw_data(data, is_test=False, tokenize_text=True, remove_no_label_data
             This is effective only when is_test=False. Defaults to False.
 
     Returns:
-        pandas.DataFrame: Data composed of index, label, and tokenized text.
+        dict: [{(optional: "index": ..., )"label": ..., "text": ...}, ...]
     """
     assert isinstance(data, str) or isinstance(data, pd.DataFrame), "Data must be from a file or pandas dataframe."
     if isinstance(data, str):
@@ -176,9 +175,7 @@ def _load_raw_data(data, is_test=False, tokenize_text=True, remove_no_label_data
 
     data["label"] = data["label"].astype(str).map(lambda s: s.split())
     if tokenize_text:
-        # multiprocessing requires serializable objects
-        with ProcessPoolExecutor() as executor:
-            data["text"] = pd.Series(tqdm(executor.map(tokenize, data["text"]), total=len(data["text"])))
+        data["text"] = data["text"].map(tokenize)
     data = data.to_dict("records")
     if not is_test:
         num_no_label_data = sum(1 for d in data if len(d["label"]) == 0)
@@ -222,15 +219,12 @@ def load_datasets(
     Returns:
         dict: A dictionary of datasets.
     """
-    if isinstance(training_data, str) or isinstance(test_data, str):
-        assert training_data or test_data, "At least one of `training_data` and `test_data` must be specified."
-    elif isinstance(training_data, pd.DataFrame) or isinstance(test_data, pd.DataFrame):
-        assert (
-            not training_data.empty or not test_data.empty
-        ), "At least one of `training_data` and `test_data` must be specified."
+    if training_data is None and test_data is None:
+        raise ValueError("At least one of `training_data` and `test_data` must be specified.")
 
     datasets = {}
     if training_data is not None:
+        logging.info(f"Loading training data")
         datasets["train"] = _load_raw_data(
             training_data, tokenize_text=tokenize_text, remove_no_label_data=remove_no_label_data
         )
@@ -243,11 +237,12 @@ def load_datasets(
         datasets["train"], datasets["val"] = train_test_split(datasets["train"], test_size=val_size, random_state=42)
 
     if test_data is not None:
+        logging.info(f"Loading test data")
         datasets["test"] = _load_raw_data(
             test_data, is_test=True, tokenize_text=tokenize_text, remove_no_label_data=remove_no_label_data
         )
 
-    if merge_train_val:
+    if merge_train_val and "val" in datasets:
         datasets["train"] = datasets["train"] + datasets["val"]
         for i in range(len(datasets["train"])):
             datasets["train"][i]["index"] = i

{libmultilabel-0.5.1 → libmultilabel-0.5.2}/libmultilabel/nn/metrics.py

@@ -217,19 +217,17 @@ def get_metrics(metric_threshold, monitor_metrics, num_classes, top_k=None):
 
         if match_top_k:
             metric_abbr = match_top_k.group(1)  # P, R, PR, or nDCG
-            top_k = int(match_top_k.group(2))
-            if top_k >= num_classes:
-                raise ValueError(
-                    f"Invalid metric: {metric}. top_k ({top_k}) is greater than num_classes({num_classes})."
-                )
+            k = int(match_top_k.group(2))
+            if k >= num_classes:
+                raise ValueError(f"Invalid metric: {metric}. k ({k}) is greater than num_classes({num_classes}).")
             if metric_abbr == "P":
-                metrics[metric] = Precision(num_classes, average="samples", top_k=top_k)
+                metrics[metric] = Precision(num_classes, average="samples", top_k=k)
             elif metric_abbr == "R":
-                metrics[metric] = Recall(num_classes, average="samples", top_k=top_k)
+                metrics[metric] = Recall(num_classes, average="samples", top_k=k)
             elif metric_abbr == "RP":
-                metrics[metric] = RPrecision(top_k=top_k)
+                metrics[metric] = RPrecision(top_k=k)
             elif metric_abbr == "nDCG":
-                metrics[metric] = NDCG(top_k=top_k)
+                metrics[metric] = NDCG(top_k=k)
                 # The implementation in torchmetrics stores the prediction/target of all batches,
                 # which can lead to CUDA out of memory.
                 # metrics[metric] = RetrievalNormalizedDCG(k=top_k)

{libmultilabel-0.5.1 → libmultilabel-0.5.2}/libmultilabel.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: libmultilabel
-Version: 0.5.1
+Version: 0.5.2
 Summary: A library for multi-label text classification
 Home-page: https://github.com/ASUS-AICS/LibMultiLabel
 Author: LibMultiLabel Team

{libmultilabel-0.5.1 → libmultilabel-0.5.2}/setup.cfg

@@ -1,6 +1,6 @@
 [metadata]
 name = libmultilabel
-version = 0.5.1
+version = 0.5.2
 author = LibMultiLabel Team
 license = MIT License
 license_file = LICENSE