use tqdm.auto for progress bars

Use tqdm.auto to display GUI progress bars when in jupyter notebooks
and console progress bars otherwise.

Also tweak travisci for python3 on OSX to hopefully run faster

Author: benfred

.travis.yml

@@ -10,20 +10,23 @@ matrix:
     - language: generic
       os: osx
       python: 2.7.14
-      osx_image: xcode9.3
       env: PYTHON=2
+      osx_image: xcode11
     - language: generic
       os: osx
       python: 3.6.5
-      osx_image: xcode9.3
       env: PYTHON=3
+      osx_image: xcode11
 
 addons:
   apt:
     packages:
     - libblas-dev
     - liblapack-dev
     - gfortran
+  homebrew:
+    packages:
+    - python3
 
 before_install:
 - |
@@ -33,8 +36,6 @@ before_install:
     PIP=pip2
     PY=python2
     if  [ "${PYTHON:0:1}" = "3" ]; then
-      brew upgrade python
-      brew install python3
       PIP=pip3
       PY=python3
     fi
@@ -63,3 +64,4 @@ cache:
   - apt
   - directories:
     - $HOME/.cache/pip
+

implicit/_als.cpp

@@ -11,7 +11,7 @@ from libcpp cimport bool
 from libcpp.vector cimport vector
 from libcpp.utility cimport pair
 
-import tqdm
+from tqdm.auto import tqdm
 
 
 cdef extern from "nearest_neighbours.h" namespace "implicit" nogil:
@@ -120,7 +120,7 @@ def all_pairs_knn(items, unsigned int K=100, int num_threads=0, show_progress=Tr
     cdef long[:] rows = np.zeros(item_count * K, dtype=int)
     cdef long[:] cols = np.zeros(item_count * K, dtype=int)
 
-    progress = tqdm.tqdm(total=item_count, disable=not show_progress)
+    progress = tqdm(total=item_count, disable=not show_progress)
     with nogil, parallel(num_threads=num_threads):
         # allocate memory per thread
         neighbours = new SparseMatrixMultiplier[int, double](item_count)

implicit/_nearest_neighbours.cpp

@@ -7,7 +7,7 @@
 import numpy as np
 import scipy
 import scipy.sparse
-import tqdm
+from tqdm.auto import tqdm
 
 import implicit.cuda
 
@@ -151,17 +151,15 @@ def fit(self, item_users, show_progress=True):
         solver = self.solver
 
         log.debug("Running %i ALS iterations", self.iterations)
-        with tqdm.tqdm(total=self.iterations, disable=not show_progress) as progress:
+        with tqdm(total=self.iterations, disable=not show_progress) as progress:
             # alternate between learning the user_factors from the item_factors and vice-versa
             for iteration in range(self.iterations):
                 s = time.time()
                 solver(Cui, self.user_factors, self.item_factors, self.regularization,
                        num_threads=self.num_threads)
-                progress.update(.5)
-
                 solver(Ciu, self.item_factors, self.user_factors, self.regularization,
                        num_threads=self.num_threads)
-                progress.update(.5)
+                progress.update(1)
 
                 if self.calculate_training_loss:
                     loss = _als.calculate_loss(Cui, self.user_factors, self.item_factors,
@@ -192,13 +190,12 @@ def _fit_gpu(self, Ciu_host, Cui_host, show_progress=True):
 
         solver = implicit.cuda.CuLeastSquaresSolver(self.factors)
         log.debug("Running %i ALS iterations", self.iterations)
-        with tqdm.tqdm(total=self.iterations, disable=not show_progress) as progress:
+        with tqdm(total=self.iterations, disable=not show_progress) as progress:
             for iteration in range(self.iterations):
                 s = time.time()
                 solver.least_squares(Cui, X, Y, self.regularization, self.cg_steps)
-                progress.update(.5)
                 solver.least_squares(Ciu, Y, X, self.regularization, self.cg_steps)
-                progress.update(.5)
+                progress.update(1)
 
                 if self.fit_callback:
                     self.fit_callback(iteration, time.time() - s)

implicit/_nearest_neighbours.pyx

@@ -4,7 +4,7 @@ import logging
 import multiprocessing
 import random
 import time
-import tqdm
+from tqdm.auto import tqdm
 
 from cython.parallel import parallel, prange
 from libc.math cimport exp
@@ -183,7 +183,7 @@ class BayesianPersonalizedRanking(MatrixFactorizationBase):
         # initialize RNG's, one per thread.
         cdef RNGVector rng = RNGVector(num_threads, len(user_items.data) - 1)
         log.debug("Running %i BPR training epochs", self.iterations)
-        with tqdm.tqdm(total=self.iterations, disable=not show_progress) as progress:
+        with tqdm(total=self.iterations, disable=not show_progress) as progress:
             for epoch in range(self.iterations):
                 correct, skipped = bpr_update(rng, userids, user_items.indices, user_items.indptr,
                                               self.user_factors, self.item_factors,
@@ -212,7 +212,7 @@ class BayesianPersonalizedRanking(MatrixFactorizationBase):
         Y = implicit.cuda.CuDenseMatrix(self.item_factors)
 
         log.debug("Running %i BPR training epochs", self.iterations)
-        with tqdm.tqdm(total=self.iterations, disable=not show_progress) as progress:
+        with tqdm(total=self.iterations, disable=not show_progress) as progress:
             for epoch in range(self.iterations):
                 correct, skipped = implicit.cuda.cu_bpr_update(userids, itemids, indptr,
                                                                X, Y, self.learning_rate,

implicit/als.py

@@ -4,7 +4,7 @@
 except ImportError:
     from urllib import urlretrieve
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 
 LOCAL_CACHE_DIR = os.path.join(os.path.expanduser("~"), "implicit_datasets")

implicit/bpr.cpp

@@ -1,6 +1,6 @@
 import h5py
 import time
-import tqdm
+from tqdm.auto import tqdm
 import os
 import logging
 from scipy.sparse import coo_matrix, csr_matrix
@@ -89,7 +89,7 @@ def _join_summary_file(data, summary_filename="msd_summary_file.h5"):
 
     # join on trackid to the summary file to get the artist/album/songname
     track_info = np.empty(shape=(len(track_lookup), 4), dtype=np.object)
-    with tqdm.tqdm(total=len(track_info)) as progress:
+    with tqdm(total=len(track_info)) as progress:
         for song in msd['metadata']['songs']:
             trackid = song[17]
             if trackid in track_lookup: