tmplot 0.2.0__tar.gz → 0.4.0__tar.gz

{tmplot-0.2.0 → tmplot-0.4.0}/MANIFEST.in

@@ -1,3 +1,4 @@
 include LICENSE
 include README.md
+include CHANGELOG.md
 recursive-exclude tests *

tmplot-0.4.0/PKG-INFO

@@ -0,0 +1,169 @@
+Metadata-Version: 2.4
+Name: tmplot
+Version: 0.4.0
+Summary: Visualization of Topic Modeling Results
+Author-email: Maksim Terpilovskii <maximtrp@gmail.com>
+License-Expression: MIT
+Project-URL: homepage, https://github.com/maximtrp/tmplot
+Project-URL: documentation, https://tmplot.readthedocs.io/
+Keywords: data science,data analytics
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Information Analysis
+Classifier: Topic :: Text Processing :: General
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.23
+Requires-Dist: pandas>=1.5
+Requires-Dist: scikit-learn>=1.4
+Requires-Dist: scipy>=1.9
+Requires-Dist: altair>=5
+Requires-Dist: ipywidgets>=8
+Requires-Dist: IPython>=8
+Requires-Dist: tqdm>=4.64
+Provides-Extra: test
+Requires-Dist: pytest; extra == "test"
+Requires-Dist: coverage; extra == "test"
+Requires-Dist: build; extra == "test"
+Requires-Dist: twine; extra == "test"
+Provides-Extra: models
+Requires-Dist: tomotopy>=0.8.0; extra == "models"
+Requires-Dist: gensim; extra == "models"
+Requires-Dist: bitermplus; extra == "models"
+Dynamic: license-file
+
+# tmplot
+
+[![Codacy coverage](https://img.shields.io/codacy/coverage/5939b1cf99bc4f9d91de11c0d3ff9e50)](https://app.codacy.com/gh/maximtrp/tmplot/coverage)
+[![Codacy grade](https://img.shields.io/codacy/grade/5939b1cf99bc4f9d91de11c0d3ff9e50)](https://app.codacy.com/gh/maximtrp/tmplot)
+[![GitHub Workflow Status](https://img.shields.io/github/actions/workflow/status/maximtrp/tmplot/python-package.yml?label=tests)](https://github.com/maximtrp/tmplot/actions/workflows/python-package.yml)
+[![Documentation Status](https://readthedocs.org/projects/tmplot/badge/?version=latest)](https://tmplot.readthedocs.io/en/latest/?badge=latest)
+[![Downloads](https://static.pepy.tech/badge/tmplot)](https://pepy.tech/project/tmplot)
+[![PyPI](https://img.shields.io/pypi/v/tmplot)](https://pypi.org/project/tmplot)
+[![Issues](https://img.shields.io/github/issues/maximtrp/tmplot.svg)](https://github.com/maximtrp/tmplot/issues)
+
+**tmplot** is a comprehensive Python package for **topic modeling analysis and visualization**. Built for data scientists and researchers, it provides powerful interactive reports and advanced analytics that extend beyond traditional LDAvis/pyLDAvis capabilities.
+
+**Analyze** • **Visualize** • **Compare** multiple topic models with ease
+
+![Plots](https://raw.githubusercontent.com/maximtrp/tmplot/main/images/topics_terms_plots.png)
+
+## Key Features
+
+### Interactive Visualization
+
+- **Topic scatter plots** with customizable coordinates and sizing
+- **Term probability charts** with relevance weighting
+- **Document analysis** showing top documents per topic
+- **Interactive reports** with real-time parameter adjustment
+
+### Advanced Analytics
+
+- **Topic stability analysis** across multiple model runs
+- **Model comparison** with sophisticated distance metrics
+- **Saliency calculations** for term importance
+- **Entropy metrics** for model optimization
+
+### Model Support
+
+- **[tomotopy](https://bab2min.github.io/tomotopy/)**: `LDAModel`, `LLDAModel`, `CTModel`, `DMRModel`, `HDPModel`, `PTModel`, `SLDAModel`, `GDMRModel`
+- **[gensim](https://radimrehurek.com/gensim/)**: `LdaModel`, `LdaMulticore`
+- **[bitermplus](https://github.com/maximtrp/bitermplus)**: `BTM`
+
+### Distance Metrics
+
+- Kullback-Leibler (symmetric & non-symmetric)
+- Jensen-Shannon divergence
+- Jeffrey's divergence
+- Hellinger & Bhattacharyya distances
+- Total variation distance
+- Jaccard index
+
+### Dimensionality Reduction
+
+t-SNE, SpectralEmbedding, MDS, LocallyLinearEmbedding, Isomap
+
+## Donate
+
+If you find this package useful, please consider donating any amount of money. This will help me spend more time on supporting open-source software.
+
+<a href="https://www.buymeacoffee.com/maximtrp" target="_blank"><img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" alt="Buy Me A Coffee" style="height: 60px !important;width: 217px !important;" ></a>
+
+## Quick Start
+
+### Installation
+
+```bash
+# From PyPI (recommended)
+pip install tmplot
+
+# Development version
+pip install git+https://github.com/maximtrp/tmplot.git
+```
+
+### Basic Usage
+
+```python
+import tmplot as tmp
+
+# Load your topic model and documents
+model = your_fitted_model  # tomotopy, gensim, or bitermplus
+docs = your_documents
+
+# Create interactive report
+tmp.report(model, docs=docs)
+
+# Or create individual visualizations
+coords = tmp.prepare_coords(model)
+tmp.plot_scatter_topics(coords, size_col='size')
+```
+
+## Advanced Examples
+
+### Get Stable Topics
+
+```python
+import tmplot as tmp
+
+# Find stable topics across multiple models
+models = [model1, model2, model3, model4]
+closest_topics, distances = tmp.get_closest_topics(models)
+stable_topics, stable_distances = tmp.get_stable_topics(closest_topics, distances)
+```
+
+### Analyze Model
+
+```python
+# Calculate entropy for model selection
+entropy_score = tmp.entropy(phi_matrix)
+
+# Analyze topic stability
+saliency = tmp.get_salient_terms(phi, theta)
+```
+
+### Visualize
+
+```python
+# Create topic distance matrix with different metrics
+topic_dists = tmp.get_topics_dist(phi, method='jsd')
+
+# Generate coordinates with custom algorithm
+coords = tmp.get_topics_scatter(topic_dists, theta, method='tsne')
+tmp.plot_scatter_topics(coords, topic=3)  # Highlight topic 3
+```
+
+## Documentation & Examples
+
+- [Complete Tutorial](https://tmplot.readthedocs.io/en/latest/tutorial.html) - Step-by-step guide
+- [API Reference](https://tmplot.readthedocs.io/) - Full documentation
+- [Tutorial Notebook](docs/source/tutorial.ipynb) - Jupyter walkthrough
+
+## Requirements
+
+**Core dependencies:** `numpy`, `scipy`, `scikit-learn`, `pandas`, `altair`, `ipywidgets`
+
+**Optional models:** `tomotopy`, `gensim`, `bitermplus`

tmplot-0.4.0/README.md

@@ -0,0 +1,131 @@
+# tmplot
+
+[![Codacy coverage](https://img.shields.io/codacy/coverage/5939b1cf99bc4f9d91de11c0d3ff9e50)](https://app.codacy.com/gh/maximtrp/tmplot/coverage)
+[![Codacy grade](https://img.shields.io/codacy/grade/5939b1cf99bc4f9d91de11c0d3ff9e50)](https://app.codacy.com/gh/maximtrp/tmplot)
+[![GitHub Workflow Status](https://img.shields.io/github/actions/workflow/status/maximtrp/tmplot/python-package.yml?label=tests)](https://github.com/maximtrp/tmplot/actions/workflows/python-package.yml)
+[![Documentation Status](https://readthedocs.org/projects/tmplot/badge/?version=latest)](https://tmplot.readthedocs.io/en/latest/?badge=latest)
+[![Downloads](https://static.pepy.tech/badge/tmplot)](https://pepy.tech/project/tmplot)
+[![PyPI](https://img.shields.io/pypi/v/tmplot)](https://pypi.org/project/tmplot)
+[![Issues](https://img.shields.io/github/issues/maximtrp/tmplot.svg)](https://github.com/maximtrp/tmplot/issues)
+
+**tmplot** is a comprehensive Python package for **topic modeling analysis and visualization**. Built for data scientists and researchers, it provides powerful interactive reports and advanced analytics that extend beyond traditional LDAvis/pyLDAvis capabilities.
+
+**Analyze** • **Visualize** • **Compare** multiple topic models with ease
+
+![Plots](https://raw.githubusercontent.com/maximtrp/tmplot/main/images/topics_terms_plots.png)
+
+## Key Features
+
+### Interactive Visualization
+
+- **Topic scatter plots** with customizable coordinates and sizing
+- **Term probability charts** with relevance weighting
+- **Document analysis** showing top documents per topic
+- **Interactive reports** with real-time parameter adjustment
+
+### Advanced Analytics
+
+- **Topic stability analysis** across multiple model runs
+- **Model comparison** with sophisticated distance metrics
+- **Saliency calculations** for term importance
+- **Entropy metrics** for model optimization
+
+### Model Support
+
+- **[tomotopy](https://bab2min.github.io/tomotopy/)**: `LDAModel`, `LLDAModel`, `CTModel`, `DMRModel`, `HDPModel`, `PTModel`, `SLDAModel`, `GDMRModel`
+- **[gensim](https://radimrehurek.com/gensim/)**: `LdaModel`, `LdaMulticore`
+- **[bitermplus](https://github.com/maximtrp/bitermplus)**: `BTM`
+
+### Distance Metrics
+
+- Kullback-Leibler (symmetric & non-symmetric)
+- Jensen-Shannon divergence
+- Jeffrey's divergence
+- Hellinger & Bhattacharyya distances
+- Total variation distance
+- Jaccard index
+
+### Dimensionality Reduction
+
+t-SNE, SpectralEmbedding, MDS, LocallyLinearEmbedding, Isomap
+
+## Donate
+
+If you find this package useful, please consider donating any amount of money. This will help me spend more time on supporting open-source software.
+
+<a href="https://www.buymeacoffee.com/maximtrp" target="_blank"><img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" alt="Buy Me A Coffee" style="height: 60px !important;width: 217px !important;" ></a>
+
+## Quick Start
+
+### Installation
+
+```bash
+# From PyPI (recommended)
+pip install tmplot
+
+# Development version
+pip install git+https://github.com/maximtrp/tmplot.git
+```
+
+### Basic Usage
+
+```python
+import tmplot as tmp
+
+# Load your topic model and documents
+model = your_fitted_model  # tomotopy, gensim, or bitermplus
+docs = your_documents
+
+# Create interactive report
+tmp.report(model, docs=docs)
+
+# Or create individual visualizations
+coords = tmp.prepare_coords(model)
+tmp.plot_scatter_topics(coords, size_col='size')
+```
+
+## Advanced Examples
+
+### Get Stable Topics
+
+```python
+import tmplot as tmp
+
+# Find stable topics across multiple models
+models = [model1, model2, model3, model4]
+closest_topics, distances = tmp.get_closest_topics(models)
+stable_topics, stable_distances = tmp.get_stable_topics(closest_topics, distances)
+```
+
+### Analyze Model
+
+```python
+# Calculate entropy for model selection
+entropy_score = tmp.entropy(phi_matrix)
+
+# Analyze topic stability
+saliency = tmp.get_salient_terms(phi, theta)
+```
+
+### Visualize
+
+```python
+# Create topic distance matrix with different metrics
+topic_dists = tmp.get_topics_dist(phi, method='jsd')
+
+# Generate coordinates with custom algorithm
+coords = tmp.get_topics_scatter(topic_dists, theta, method='tsne')
+tmp.plot_scatter_topics(coords, topic=3)  # Highlight topic 3
+```
+
+## Documentation & Examples
+
+- [Complete Tutorial](https://tmplot.readthedocs.io/en/latest/tutorial.html) - Step-by-step guide
+- [API Reference](https://tmplot.readthedocs.io/) - Full documentation
+- [Tutorial Notebook](docs/source/tutorial.ipynb) - Jupyter walkthrough
+
+## Requirements
+
+**Core dependencies:** `numpy`, `scipy`, `scikit-learn`, `pandas`, `altair`, `ipywidgets`
+
+**Optional models:** `tomotopy`, `gensim`, `bitermplus`

{tmplot-0.2.0 → tmplot-0.4.0}/pyproject.toml

@@ -1,5 +1,5 @@
 [build-system]
-requires = ["setuptools>=61.0", "wheel"]
+requires = ["setuptools>=77.0", "wheel"]
 build-backend = "setuptools.build_meta"
 
 [project]
@@ -7,8 +7,9 @@ name = "tmplot"
 dynamic = ["version"]
 description = "Visualization of Topic Modeling Results"
 readme = "README.md"
-requires-python = ">=3.8"
-license.file = "LICENSE"
+requires-python = ">=3.9"
+license = "MIT"
+license-files = ["LICENSE"]
 authors = [
   { name = "Maksim Terpilovskii", email = "maximtrp@gmail.com" },
 ]
@@ -17,9 +18,7 @@ keywords = [
     "data analytics"
 ]
 classifiers = [
-    "License :: OSI Approved :: MIT License",
     "Operating System :: OS Independent",
-    "Programming Language :: Python :: 3.8",
     "Programming Language :: Python :: 3.9",
     "Programming Language :: Python :: 3.10",
     "Programming Language :: Python :: 3.11",
@@ -31,18 +30,18 @@ urls.homepage = "https://github.com/maximtrp/tmplot"
 urls.documentation = "https://tmplot.readthedocs.io/"
 
 dependencies = [
-    "numpy",
-    "seaborn",
-    "pandas",
-    "scikit-learn",
-    "scipy",
-    "altair",
-    "ipywidgets",
-    "tqdm",
+    "numpy>=1.23",
+    "pandas>=1.5",
+    "scikit-learn>=1.4",
+    "scipy>=1.9",
+    "altair>=5",
+    "ipywidgets>=8",
+    "IPython>=8",
+    "tqdm>=4.64",
 ]
 
 [project.optional-dependencies]
-test = ["pytest", "coverage"]
+test = ["pytest", "coverage", "build", "twine"]
 models = ["tomotopy>=0.8.0", "gensim", "bitermplus"]
 
 [tool.setuptools.dynamic]

{tmplot-0.2.0 → tmplot-0.4.0}/src/tmplot/__init__.py

@@ -5,4 +5,4 @@ from ._stability import *  # noqa: F401, F403
 from ._vis import *  # noqa: F401, F403
 from ._metrics import *  # noqa: F401, F403
 
-__version__ = '0.2.0'
+__version__ = "0.4.0"

{tmplot-0.2.0 → tmplot-0.4.0}/src/tmplot/_distance.py

@@ -1,5 +1,6 @@
 __all__ = ["get_topics_dist", "get_topics_scatter", "get_top_topic_words"]
 from typing import Optional, Union, List
+from inspect import signature
 from itertools import combinations
 from pandas import DataFrame, Index
 import numpy as np
@@ -15,37 +16,48 @@ from sklearn.manifold import (
 from ._helpers import calc_topics_marg_probs
 
 
+EPSILON = 1e-64
+
+
+def _positive_probabilities(values: np.ndarray) -> np.ndarray:
+    values = np.clip(np.asarray(values, dtype=float), EPSILON, None)
+    return values / values.sum()
+
+
 def _dist_klb(a1: np.ndarray, a2: np.ndarray):
-    dist = kl_div(a1, a2)
-    return dist[np.isfinite(dist)].sum()
+    return kl_div(_positive_probabilities(a1), _positive_probabilities(a2)).sum()
 
 
 def _dist_sklb(a1: np.ndarray, a2: np.ndarray):
-    dist = kl_div(a1, a2) + kl_div(a1, a2)
-    return dist[np.isfinite(dist)].sum()
+    a1_safe = _positive_probabilities(a1)
+    a2_safe = _positive_probabilities(a2)
+    return (kl_div(a1_safe, a2_safe) + kl_div(a2_safe, a1_safe)).sum()
 
 
 def _dist_jsd(a1: np.ndarray, a2: np.ndarray):
-    dist = 0.5 * kl_div(a1, a2) + 0.5 * kl_div(a1, a2)
+    m = 0.5 * (a1 + a2)
+    dist = 0.5 * kl_div(a1, m) + 0.5 * kl_div(a2, m)
     return dist[np.isfinite(dist)].sum()
 
 
 def _dist_jef(a1: np.ndarray, a2: np.ndarray):
-    vals = (a1 - a2) * (np.log(a1) - np.log(a2))
-    vals[(vals <= 0) | ~np.isfinite(vals)] = 0.0
-    return vals.sum()
+    a1_safe = _positive_probabilities(a1)
+    a2_safe = _positive_probabilities(a2)
+    return ((a1_safe - a2_safe) * (np.log(a1_safe) - np.log(a2_safe))).sum()
 
 
 def _dist_hel(a1: np.ndarray, a2: np.ndarray):
-    a1[(a1 <= 0) | ~np.isfinite(a1)] = 1e-64
-    a2[(a2 <= 0) | ~np.isfinite(a2)] = 1e-64
-    hel_val = distance.euclidean(np.sqrt(a1), np.sqrt(a2)) / np.sqrt(2)
+    a1_safe = a1.copy()
+    a2_safe = a2.copy()
+    a1_safe[(a1_safe <= 0) | ~np.isfinite(a1_safe)] = EPSILON
+    a2_safe[(a2_safe <= 0) | ~np.isfinite(a2_safe)] = EPSILON
+    hel_val = distance.euclidean(np.sqrt(a1_safe), np.sqrt(a2_safe)) / np.sqrt(2)
     return hel_val
 
 
 def _dist_bhat(a1: np.ndarray, a2: np.ndarray):
     pq = a1 * a2
-    pq[(pq <= 0) | ~np.isfinite(pq)] = 1e-64
+    pq[(pq <= 0) | ~np.isfinite(pq)] = EPSILON
     dist = -np.log(np.sum(np.sqrt(pq)))
     return dist
 
@@ -92,6 +104,14 @@ def get_topics_dist(
         Topics distances matrix.
     """
     phi_copy = np.array(phi)
+
+    if phi_copy.ndim != 2:
+        raise ValueError("phi must be a 2D array (words x topics)")
+    if np.any(phi_copy < 0):
+        raise ValueError("phi must contain non-negative probability values")
+    if not np.allclose(phi_copy.sum(axis=0), 1.0, atol=1e-6):
+        raise ValueError("phi columns must sum to 1 (probability distributions)")
+
     topics_num = phi_copy.shape[1]
     topics_pairs = combinations(range(topics_num), 2)
 
@@ -109,8 +129,12 @@ def get_topics_dist(
         "jac": _dist_jac,
     }
 
+    if method not in dist_funcs:
+        raise ValueError(
+            f"Unknown distance method {method!r}; choose from {sorted(dist_funcs)}"
+        )
+    _dist_func = dist_funcs[method]
     for i, j in topics_pairs:
-        _dist_func = dist_funcs.get(method, "sklb")
         topics_dists[((i, j), (j, i))] = _dist_func(
             phi_copy[:, i], phi_copy[:, j], **kwargs
         )
@@ -118,6 +142,17 @@ def get_topics_dist(
     return topics_dists
 
 
+def _classical_mds(distances: np.ndarray) -> np.ndarray:
+    count = distances.shape[0]
+    centering = np.eye(count) - np.ones((count, count)) / count
+    gram = -0.5 * centering @ (distances**2) @ centering
+    eigenvalues, eigenvectors = np.linalg.eigh(gram)
+    positive = eigenvalues > np.finfo(float).eps
+    if not positive.any():
+        return np.zeros((count, 1))
+    return eigenvectors[:, positive] * np.sqrt(eigenvalues[positive])
+
+
 def get_topics_scatter(
     topic_dists: np.ndarray,
     theta: np.ndarray,
@@ -149,11 +184,39 @@ def get_topics_scatter(
     DataFrame
         Topics scatter coordinates.
     """
-    if not method_kws:
-        method_kws = {"n_components": 2}
+    topic_dists = np.asarray(topic_dists, dtype=float)
+    theta = np.asarray(theta, dtype=float)
+    if topic_dists.ndim != 2 or topic_dists.shape[0] != topic_dists.shape[1]:
+        raise ValueError("topic_dists must be a square 2D matrix")
+    if not np.isfinite(topic_dists).all():
+        raise ValueError("topic_dists must contain only finite values")
+    if not np.allclose(topic_dists, topic_dists.T):
+        raise ValueError("topic_dists must be symmetric")
+    if theta.ndim != 2 or theta.shape[0] != topic_dists.shape[0]:
+        raise ValueError("theta topics dimension must match topic_dists")
+    if topic_dists.shape[0] < 2:
+        raise ValueError("at least two topics are required for scatter coordinates")
+
+    valid_methods = ["tsne", "sem", "mds", "lle", "ltsa", "isomap"]
+    if method not in valid_methods:
+        raise ValueError(
+            f"Unknown scatter method {method!r}; choose from {valid_methods}"
+        )
+
+    if topic_dists.shape[0] == 2:
+        half_distance = topic_dists[0, 1] / 2
+        coords = np.array([[-half_distance, 0.0], [half_distance, 0.0]])
+        topics_xy = DataFrame(coords, columns=Index(["x", "y"]))
+        topics_xy["topic"] = topics_xy.index.astype(int)
+        topics_xy["size"] = calc_topics_marg_probs(theta) * 100
+        return topics_xy
+
+    method_kws = dict(method_kws or {})
+    method_kws.setdefault("n_components", 2)
 
     if method == "tsne":
-        method_kws.setdefault("init", "pca")
+        method_kws.setdefault("metric", "precomputed")
+        method_kws.setdefault("init", "random")
         method_kws.setdefault("learning_rate", "auto")
         method_kws.setdefault("perplexity", min(50, max(topic_dists.shape[0] // 2, 1)))
         transformer = TSNE(**method_kws)
@@ -161,24 +224,46 @@ def get_topics_scatter(
     elif method == "sem":
         method_kws.setdefault("affinity", "precomputed")
         transformer = SpectralEmbedding(**method_kws)
+        nonzero = topic_dists[topic_dists > 0]
+        scale = np.median(nonzero) if nonzero.size else 1.0
+        transform_input = np.exp(-((topic_dists / scale) ** 2))
+        np.fill_diagonal(transform_input, 1.0)
 
     elif method == "mds":
-        method_kws.setdefault("dissimilarity", "precomputed")
+        mds_params = signature(MDS.__init__).parameters
+        if "metric_mds" in mds_params:
+            # scikit-learn >= 1.9 deprecated `dissimilarity` in favor of `metric`
+            method_kws.setdefault("metric", "precomputed")
+        else:
+            method_kws.setdefault("dissimilarity", "precomputed")
         method_kws.setdefault("normalized_stress", "auto")
+        method_kws.setdefault("n_init", 1)
+        if "init" in mds_params:
+            # `init` is a constructor argument only since scikit-learn 1.9
+            method_kws.setdefault("init", "random")
         transformer = MDS(**method_kws)
 
     elif method == "lle":
         method_kws["method"] = "standard"
+        method_kws.setdefault("n_neighbors", min(5, topic_dists.shape[0] - 1))
         transformer = LocallyLinearEmbedding(**method_kws)
+        transform_input = _classical_mds(topic_dists)
 
     elif method == "ltsa":
         method_kws["method"] = "ltsa"
+        method_kws.setdefault("n_neighbors", min(5, topic_dists.shape[0] - 1))
         transformer = LocallyLinearEmbedding(**method_kws)
+        transform_input = _classical_mds(topic_dists)
 
     elif method == "isomap":
+        method_kws.setdefault("metric", "precomputed")
+        method_kws.setdefault("n_neighbors", min(5, topic_dists.shape[0] - 1))
         transformer = Isomap(**method_kws)
 
-    coords = transformer.fit_transform(topic_dists)
+    else:
+        raise AssertionError("validated scatter method was not handled")
+
+    coords = transformer.fit_transform(locals().get("transform_input", topic_dists))
 
     topics_xy = DataFrame(coords, columns=Index(["x", "y"]))
     topics_xy["topic"] = topics_xy.index.astype(int)
@@ -213,6 +298,7 @@ def get_top_topic_words(
     DataFrame
         Words with highest probabilities in all (or selected) topics.
     """
-    return phi.loc[:, topics_idx or phi.columns].apply(
+    selected_topics = phi.columns if topics_idx is None else topics_idx
+    return phi.loc[:, selected_topics].apply(
         lambda x: x.sort_values(ascending=False).head(words_num).index, axis=0
     )