Merge pull request YingfanWang#78 from YingfanWang/hotfix

Release 0.7.4

demo/basic_demo.py

@@ -12,9 +12,9 @@
 # Initialize the pacmap instance
 # By default, the n_neighbors is set to 10.
 # Setting n_neighbors to "None" can enable an automatic parameter selection
-# choice shown in "parameter" section of the README file. 
+# choice shown in "parameter" section of the README file.
 # Notice that from v0.6.0 on, we rename the n_dims parameter to n_components.
-reducer = pacmap.PaCMAP(n_components=2, n_neighbors=10, MN_ratio=0.5, FP_ratio=2.0) 
+reducer = pacmap.PaCMAP(n_components=2, n_neighbors=10, MN_ratio=0.5, FP_ratio=2.0)
 
 # fit the data (The index of transformed data corresponds to the index of the original data)
 X_transformed = reducer.fit_transform(X, init="pca")

demo/demo_utils.py

@@ -1,18 +1,20 @@
 import matplotlib.pyplot as plt
 
+
 def generate_figure(embedding, labels, title):
     fig, ax = plt.subplots(1, 1, figsize=(6, 6))
     ax.scatter(embedding[:, 0], embedding[:, 1], s=0.5, c=labels, cmap='Spectral')
     ax.axis('off')
     ax.set_title(title)
     plt.savefig(f"./{title}.png")
 
+
 def generate_combined_figure(embeddings, labels, titles, theme_title):
     len_subfigs = len(embeddings)
     assert len(labels) == len_subfigs
     assert len(titles) == len_subfigs
     n_rows = (len_subfigs + 2) // 3
-    fig, axes = plt.subplots(n_rows, 3, figsize = (18, n_rows * 6))
+    fig, axes = plt.subplots(n_rows, 3, figsize=(18, n_rows * 6))
     axes = axes.flatten()
     for i in range(len_subfigs):
         ax = axes[i]

demo/specify_nn_demo.py

@@ -23,10 +23,10 @@
 
 nbrs = np.zeros((n, 20), dtype=np.int32)
 for i in range(n):
-    nbrs_ = tree.get_nns_by_item(i, 20 + 1) # The first nbr is always the point itself
+    nbrs_ = tree.get_nns_by_item(i, 20 + 1)  # The first nbr is always the point itself
     nbrs[i, :] = nbrs_[1:]
 
-scaled_dist = np.ones((n, n_neighbors)) # No scaling is needed
+scaled_dist = np.ones((n, n_neighbors))  # No scaling is needed
 
 # Type casting is needed for numba acceleration
 X = X.astype(np.float32)
@@ -37,7 +37,11 @@
 
 # initializing the pacmap instance
 # feed the pair_neighbors into the instance
-embedding = pacmap.PaCMAP(n_components=2, n_neighbors=n_neighbors, MN_ratio=0.5, FP_ratio=2.0, pair_neighbors=pair_neighbors) 
+embedding = pacmap.PaCMAP(n_components=2,
+                          n_neighbors=n_neighbors,
+                          MN_ratio=0.5,
+                          FP_ratio=2.0,
+                          pair_neighbors=pair_neighbors)
 
 # fit the data (The index of transformed data corresponds to the index of the original data)
 X_transformed = embedding.fit_transform(X, init="pca")

demo/transform_demo.py

@@ -9,7 +9,7 @@
 import pacmap
 import numpy as np
 from sklearn.model_selection import StratifiedKFold
-from demo_utils import *
+import demo_utils
 
 
 # MNIST
@@ -25,7 +25,7 @@
         X_train, X_test = mnist[train_index], mnist[test_index]
         y_train, y_test = labels[train_index], labels[test_index]
         break
-    
+
     # Initialize the instance
     reducer = pacmap.PaCMAP(n_components=2, n_neighbors=10, MN_ratio=0.5, FP_ratio=2.0, random_state=20, save_tree=False)
 
@@ -41,4 +41,4 @@
     embeddings = [embedding, embedding_test, embedding_combined]
     labelset = [y_train, y_test, y]
     titles = ['Training', 'Test', 'Combined']
-    generate_combined_figure(embeddings, labelset, titles, f'mnist_transform_{n}')
+    demo_utils.generate_combined_figure(embeddings, labelset, titles, f'mnist_transform_{n}')

evaluation/evaluation.py

@@ -1,3 +1,5 @@
+# pylint: skip-file
+# flake8: noqa
 import os
 import json
 import matplotlib.cm as cm

experiments/FA2.py

@@ -1,3 +1,5 @@
+# pylint: skip-file
+# flake8: noqa
 import os
 import json
 import numpy as np
@@ -7,6 +9,7 @@
 from fa2 import ForceAtlas2 as FA2
 from sklearn.preprocessing import scale
 from sklearn.decomposition import PCA
+from sklearn.neighbors import NearestNeighbors
 
 def transform_by_FA2(X, n_neighbors=6):
     if X.shape[1] > 100:

experiments/run_experiments.py

@@ -1,3 +1,5 @@
+# pylint: skip-file
+# flake8: noqa
 import umap
 import trimap
 import FlowCal

experiments/run_experiments_LargeVis.py

@@ -1,3 +1,5 @@
+# pylint: skip-file
+# flake8: noqa
 import FlowCal
 import pandas as pd
 

experiments/run_experiments_tSNE.py

@@ -1,3 +1,5 @@
+# pylint: skip-file
+# flake8: noqa
 import FlowCal
 import json
 

rainbow_plot/rainbow_plots_for_good_loss.py

@@ -1,9 +1,8 @@
+# pylint: skip-file
+# flake8: noqa
 import matplotlib.pyplot as plt
 import numpy as np
-import umap
 from scipy import integrate
-from pacmap import PaCMAP
-from experiments.run_experiments import data_prep
 
 
 cmap_fig = plt.cm.get_cmap("Spectral")
@@ -20,56 +19,66 @@ def ladder_map(grids, ladder_range):
     l_map /= len(ladder_range)
     return l_map
 
+
 # parameter "a" and "b" use default values as below
 def attr(x):
     return -pow(x, 0.79)/(1 + pow(x, 2))
 
+
 def repul(x):
     return 0.895 * x/(1 + pow(x, 2))/(0.001 + pow(x, 2))
 
+
 def integ_attr(b):
     res = np.zeros(b.shape)
     for i in range(b.shape[1]):
         res[0][i] = integrate.quad(attr, 0, b[0][i], points=[0])[0]
     return res
 
+
 def integ_repul(b):
     res = np.zeros(b.shape)
     for i in range(b.shape[0]):
         res[i][0] = integrate.quad(repul, 0, b[i][0], points=[0])[0]
     return res
 
+
 # For t-SNE we choose a neighbor and further point to visualize forces on them (using COIL20 dataset, 300 iterations)
 def t_attr(x):
     qij = 1.0 / (x ** 2 + 1.0) / 11500
     qij = np.maximum(qij, 1e-12)
     force = - (8.58 * 1e-5 - qij) * x / (1.0 + x ** 2)
     return force
 
+
 def t_repul(x):
     qij = 1.0 / (x ** 2 + 1.0) / 11500
     qij = np.maximum(qij, 1e-12)
     force = - 10 * (1.19 * 1e-8 - qij) * x / (1.0 + x ** 2)
     return force
 
+
 def t_integ_attr(b):
     res = np.zeros(b.shape[0])
     for i in range(b.shape[0]):
         res[i] = integrate.quad(t_attr, 0, b[i], points=[0])[0]
     return res
 
+
 def t_integ_repul(b):
     res = np.zeros(b.shape[0])
     for i in range(b.shape[0]):
         res[i] = integrate.quad(t_repul, 0, b[i], points=[0])[0]
     return res
 
+
 def t_integ_attr_(b):
     res = np.zeros(b.shape)
     for i in range(b.shape[1]):
         res[0][i] = integrate.quad(t_attr, 0, b[0][i], points=[0])[0]
     return res
 
+
 def t_integ_repul_(b):
     res = np.zeros(b.shape)
     for i in range(b.shape[0]):
@@ -79,8 +88,6 @@ def t_integ_repul_(b):
 
 plt.figure(figsize=(28, 15))
 
-
-
 plt.axes([0.047, 0.52, 0.2, 0.44])
 x = np.linspace(0.0001, 100, num=7000)# d_ij
 y = np.linspace(0.0001, 100, num=7000)# d_ik
@@ -115,8 +122,6 @@ def t_integ_repul_(b):
 plt.xlabel(r'$d_{ij}$', fontsize=38)
 plt.ylabel(r'$d_{ik}$', fontsize=38)
 
-
-
 plt.axes([0.293, 0.52, 0.2, 0.44])
 x = np.linspace(0.0001, 25, num=7000) # d_ij
 y = np.linspace(0.0001, 25, num=7000) # d_ik
@@ -150,7 +155,6 @@ def t_integ_repul_(b):
 plt.xlabel(r'$d_{ij}$', fontsize=38)
 plt.ylabel(r'$d_{ik}$', fontsize=38)
 
-
 plt.axes([0.543, 0.52, 0.2, 0.44])
 x = np.linspace(0.0001, 200, num=7000) # d_ij
 y = np.linspace(0.0001, 200, num=7000) # d_ik
@@ -184,7 +188,6 @@ def t_integ_repul_(b):
 plt.xlabel(r'$d_{ij}$', fontsize=38)
 plt.ylabel(r'$d_{ik}$', fontsize=38)
 
-
 plt.axes([0.795, 0.52, 0.2, 0.44])
 x = np.linspace(0.0001, 50, num=7000) # d_ij
 y = np.linspace(0.0001, 50, num=7000) # d_ik
@@ -218,5 +221,4 @@ def t_integ_repul_(b):
 plt.xlabel(r'$d_{ij}$', fontsize=38)
 plt.ylabel(r'$d_{ik}$', fontsize=38)
 
-
 plt.savefig('rainbow_good_loss')

release_notes.md

@@ -1,5 +1,9 @@
 # Release Notes
 
+- 0.7.4
+  Fix the problem that `sample_FP_pair` and `sample_FP_pair_deterministic` may create
+  FP pairs on NN.
+
 - 0.7.3
 
   Fix the problem when the sample size is smaller than the number of neighbors,

setup.cfg

@@ -1,6 +1,6 @@
 [metadata]
 name = pacmap
-version = 0.7.3
+version = 0.7.4
 author = Yingfan Wang, Haiyang Huang, Cynthia Rudin, Yaron Shaposhnik
 author_email = yingfan.wang@duke.edu
 description = The official implementation for PaCMAP: Pairwise Controlled Manifold Approximation Projection

source/pacmap/__init__.py

@@ -1,4 +1,4 @@
-from .pacmap import *
+from .pacmap import PaCMAP, sample_neighbors_pair
 
 import pkg_resources
 __version__ = pkg_resources.get_distribution('pacmap').version