audit-trail:

• notebook-5.6 (security fix)
• scikit-learn-0.19.2 + pythonnet-2.40.dev0 +clrmagic for Python-3.7
• MLxtend (for now)
• dowgrade to pillow-5.1 until bokeh removes the constraint
• bqplot-0.11.0 + ipywidgets-7.3.1 + widgetsnbextension-3.3.1 (bracing for jupyterlab-0.33.1)
• Torch is now way bigger wheel then Tensorflow... removing Torch + will try to fit in Tensorflow again
• Jupyterlab-0.33.1 try: bqplot problem, altair API changed (simplified), adjust to changing seaborn API
• update winpython-checker.ipynb for seaborn + altair +pythonnet
• plotnine-3.1.0dev0 but remove geopandas dependancy
• winpython_checker.ipynb: adjusted to changing seaborn+altair API
• Jupyterlab-0.33.2: fix build system that was using old @jupyter-widgets/jupyterlab-manager@0.35 (many thanks Jason Grout, curiously the error allowed temporarly the holoviews 'outliers' come back in Python-3.7 64 bit)
• Jupyterlab-0.33.4
• IPython-6.5 fix for python-3.7,
• Jupyterlab_Bokeh compatibility fix for Jupyterlab-0.33.4
• usa has a back-to-school august 19th

beta1 (2018-07-31: python-3.7.0 fixes, take the Jupyterlab-0.33.x plunge, go back to Tensorflow, holoviews not working well in Jupyterlab ... a true beta )

• Jupyterlab-0.33.8
• bokeh-1.0.0dev4 (that or other makes holoviews compatible with Jupyterlab-0.33.x)
• plotnine-3.1.0 (with geopandas)
• spyder-3.3.1dev0-20180810
• idlex-1.18 (now a separate icon, as it's using a forked idle of python-3.5)
• bokeh-1.0.0dev5

beta2 (2018-08-12: stabilized Jupyterlab-beta3, brings back IDLEX)

• Spyder-3.3.1,
• Jupyterlab-0.33.11, Qtconsole-4.4.0
• Altair-2.2.1
• remove all complement "jupyter install" of vega module (now inside jupyter note book & lab)
• minesweeper.py from young jakevdp
• Altair-2.2.2
• nodejs-v8.11.4
• tensorflow_probability + Tensorflow-1.10.1 (for Python-3.6 64 bit only at the moment)
• PyQtgraph-0.11.0dev0-20180819, Holoviews-1.11.0a3
• Jupyterlab-0.34.0 (beta4 requires a quick re-spin of pyviz_comms (and maybe one day of jupyterlab_bokeh)

beta3 (2018-08-20: updated javascript stack)

• Jupyterlab_bokeh-0.6.1
• Pillow-5.2.0
• xlwings-0.11.8 , vega-1.4.0
• Jupyterlab-0.34.1
• pdvega-0.2.1.dev0-20180824 (rewritten using altair parts, the old one is not working anymore)
• bokeh-1.0.0dev6
• drop examples on pymc3,
• rise (in jupyter classic notebook only)
• removal of pymc3: dead slow because Theano doesn't compile
• ppci : WASM pure alpha awesomness, presented via a compacted form of https://github.com/almarklein/rocket_rust_py
• scikit-learn-0.20.0rc1

rc (2018-08-28: idlex, Spyder-3.3.1, Jupyterlab-0.34 beta4)

rc2 (2018-09-02: idlex was missing in the big wheels, re-spinning with IDLEX added)

• replace edward per Tensorflow-probability (if Tensorflow-1.10 on Python-3.7), and maybe drop
  pymc3+Theano next cycle (alternative is pytorch+pyro https://medium.com/@alexrachnog/financial-forecasting-with-probabilistic-programming-and-pyro-db68ab1a1dba)
• bokeh-1.0 , smells september end,
• Python-3.7.1 seems too late for this release
• Matplotlib-3.0.0rc2 doesn't fit with winpython_checker (seaborn crash + holoviews warnings)
• not sure trio: would require asyncio-trio ... nurseries and cancel scopes will be added to standard asyncio in python-3.8
• dm-sonnet is linux/mac only
• papermill looks nice but biased to AWS, waiting

build1 2018-07-31 (Jupyterlab-0.33.4 beta3, Python-3.7 fixes, pythonnet for Python-3.7)

Changes from WinPython 2018-02:

• New Packages:
  • TensorFlow replaces back Pytorch (PyTorch wheel has grown larger then Tensorflow),
  • descartes (because of plotnine)
• Upgraded Packages:
  • Python-3.7 fixes: astroid-2.0.1, ipython-6.5.0, scikit_learn-0.19.2, 2.4.0.dev0,
  • Jupyterlab-0.33.4 (beta3) and friends: ipyleaflet, bqplot,

Areas of particular interest for testers:

• Python-3.7 stack now completed,
• Jupyterlab-0.33.4, Tensorflow,
• Holoviews is not currently compatible with Jupyterlab-0.33.4, use it classic notebook only.

focus for next build:

• Python-3.7.0rc1,
• Spyder-3.3.1, Altair-2.2.0
• Holoviews compatibility with Jupyterlab-0.33.4,
• checking the outlier bug in Python-3.7 64bit

History of changes for WinPython-64bit 3.7.0.2 build1

The following changes were made to WinPython-64bit distribution since version 3.7.0.1.

Python packages

New packages:

• absl_py 0.3.0 (Abseil Python Common Libraries)

• astor 0.7.1 (Read/rewrite/write Python ASTs)

• clrmagic 0.0.1a2 (IPython cell magic to use .NET languages)

• cyordereddict 1.0.0 (Cython implementation of Python's collections.OrderedDict)

• descartes 1.1.0 (Use geometric objects as matplotlib paths and patches)

• gast 0.2.0 (Python AST that abstracts the underlying Python version)

• grpcio 1.13.0 (HTTP/2-based RPC framework)

• mlxtend 0.13.0 (Machine Learning Library Extensions)

• prometheus_client 0.3.0 ()

• protobuf 3.6.0 (Protocol Buffers - Google's data interchange format)

• pythonnet 2.4.0.dev0 (.Net and Mono integration for Python)

• tensorboard 1.9.0 (TensorBoard lets you watch Tensors Flow)

• tensorflow 1.9.0 (TensorFlow helps the tensors flow)

• termcolor 1.1.0 (ANSII Color formatting for output in terminal)

• vega_datasets 0.5.0 (A Python package for offline access to Vega datasets)

Upgraded packages:

• astroid 1.6.5 → 2.0.1 (Rebuild a new abstract syntax tree from Python's ast (required for pylint))

• beautifulsoup4 4.6.0 → 4.6.1 (Screen-scraping library)

• bqplot 0.10.5 → 0.11.0 (Interactive plotting for the Jupyter notebook, using d3.js and ipywidgets.)

• comtypes 1.1.4 → 1.1.6 (Pure Python COM package)

• cython 0.28.3 → 0.28.4 (Cython is a language that makes writing C extensions for the Python language as easy as Python)

• dask 0.18.1 → 0.18.2 (Minimal task scheduling abstraction)

• fast_histogram 0.4 → 0.5 (Fast 1D and 2D histogram functions in Python)

• geopy 1.14.0 → 1.16.0 (Python Geocoding Toolbox)

• greenlet 0.4.13 → 0.4.14 (Lightweight in-process concurrent programming)

• hvplot 0.2.0 → 0.2.1 (A high-level plotting API for pandas, dask, streamz and xarray built on HoloViews)

• ipympl 0.2.0 → 0.2.1 (Matplotlib Jupyter Extension)

• ipython 6.4.0 → 6.5.0 (Enhanced Python shell)

• ipywidgets 7.2.1 → 7.3.1 (IPython HTML widgets for Jupyter)

• joblib 0.12.0 → 0.12.1 (Lightweight pipelining: using Python functions as pipeline jobs.)

• jupyter_sphinx 0.1.2 → 0.1.3 (Jupyter Sphinx Extensions)

• jupyterlab 0.32.1 → 0.33.4 (Jupyter lab environment notebook server extension)

• jupyterlab_launcher 0.10.5 → 0.11.2 (Jupyter Launcher)

• keras 2.2.0 → 2.2.2 (Theano-based Deep Learning library)

• keras_applications 1.0.2 → 1.0.4 (Reference implementations of popular deep learning models)

• keras_preprocessing 1.0.1 → 1.0.2 (Easy data preprocessing and data augmentation for deep learning models)

• msgpack_numpy 0.4.3 → 0.4.3.1 (Numpy data serialization using msgpack)

• mypy 0.610 → 0.620 (Optional static typing for Python)

• mysql_connector_python 8.0.6 → 8.0.11 (MySQL driver written in Python)

• notebook 5.5.0 → 5.6.0 (# Jupyter Notebook)

• numexpr 2.6.5 → 2.6.6 (Fast evaluation of array expressions elementwise by using a vector-based virtual machine)

• parambokeh 0.2.2 → 0.2.3 (Declarative Python programming using Parameters.)

• paramnb 2.0.2 → 2.0.4 (Generate ipywidgets from Parameterized objects in the notebook)

• parso 0.3.0 → 0.3.1 (A Python Parser)

• pillow 5.2.0 → 5.1.1 (Python Imaging Library (fork))

• pip 10.0.1 → 18.0 (A tool for installing and managing Python packages)

• plotnine 0.3.0 → 0.3.1.dev0 (A grammar of graphics for python)

• pluggy 0.6.0 → 0.7.1 (plugin and hook calling mechanisms for python)

• pyct 0.4.3 → 0.4.5 (python package common tasks for users (e.g. copy examples, fetch data, ...))

• pygame 1.9.3 → 1.9.4 (Pygame gives multimedia to python.)

• pylint 1.9.2 → 2.0.1 (Logilab code analysis module: analyzes Python source code looking for bugs and signs of poor quality)

• pymc3 3.4.1 → 3.5 (Markov Chain Monte Carlo sampling toolkit.)

• pymongo 3.7.0 → 3.7.1 (Python driver for MongoDB http://www.mongodb.org)

• pytest 3.6.3 → 3.6.4 (pytest: simple powerful testing with Python)

• pyzmq 17.0.0 → 17.1.0 (Lightweight and super-fast messaging based on ZeroMQ library (required for IPython Qt console))

• regex 2018.6.21 → 2018.7.11 (Alternative regular expression module, to replace re.)

• reportlab 3.4.0 → 3.5.2 (The PDF generation library)

• ruamel.yaml 0.15.42 → 0.15.44 (a YAML parser/emitter that supports roundtrip preservation of comments, seq/map flow style, and map key order)

• scikit_learn 0.19.1 → 0.19.2 (A set of Python modules for machine learning and data mining)

• scs 1.2.7 → 2.0.2 (scs: splitting conic solver)

• seaborn 0.9.dev0 → 0.9.0 (statistical data visualization)

• setuptools 39.2.0 → 40.0.0 (Download, build, install, upgrade, and uninstall Python packages - easily)

• sphinx 1.7.5 → 1.7.6 (Tool for generating documentation which uses reStructuredText as its markup language)

• sphinx_rtd_theme 0.4.0 → 0.4.1 (ReadTheDocs.org theme for Sphinx, 2013 version.)

• sqlalchemy 1.2.9 → 1.2.10 (SQL Toolkit and Object Relational Mapper)

• sympy 1.1.1 → 1.2 (Symbolic Mathematics Library)

• tornado 5.0.2 → 5.1 (Scalable, non-blocking web server and tools (required for IPython notebook))

• tqdm 4.23.4 → 4.24.0 (A Simple Python Progress Meter)

• widgetsnbextension 3.2.1 → 3.3.1 (IPython HTML widgets for Jupyter)

• wordcloud 1.4.1 → 1.5.0 (A little word cloud generator)

• xarray 0.10.7 → 0.10.8 (N-D labeled arrays and datasets in Python)

Removed packages:

• torch 0.4.0 (a deep learning framework.)

• torchvision 0.2.1 (Datasets, Transforms and Models specific to Computer Vision)

Please leave PyTorch + TorchVision inside.
Those are Python written Deep Learning while TensorFlow isn't.
Moreover PyTorch is much better and nicer to users.

I think once you have PyTorch you can remove Keras, Theano, etc...
PyTorch is a must.

build2 2018-08-11 (Jupyterlab-0.33.4 beta8, Python-3.7 fixes, IDLEX)

Changes from WinPython 2018-03 build 1:

• New Packages:
  • IDLEX is back,
  • geopandas and dependancies (shapely, pyproj, fiona, munch, cligj, ...)
• Upgraded Packages:
  • Jupyterlab-0.33.4 (beta8), bokeh-1.0.0.dev5, ipywidgets-7.4.0
  • plotnine-0.4.0, matplotlib-2.2.3, pandas-0.23.4, Cython-0.28.5
  • Spyder-3.3.1dev0-20180810

Areas of particular interest for testers:

• overal stability of Jupyterlab-0.33.8 under Python-3.7.0
• IDLEX icon.

focus for next build:

• Spyder-3.3.1 official, Altair-2.2.0, bokeh-1.0.0
• may be an rc, if no news on Python-3.7.1 release date

History of changes for WinPython-64bit 3.7.0.2 b2

The following changes were made to WinPython-64bit distribution since version 3.7.0.1.

Python packages

New packages:

• absl_py 0.3.0 (Abseil Python Common Libraries)

• astor 0.7.1 (Read/rewrite/write Python ASTs)

• click_plugins 1.0.3 (An extension module for click to enable registering CLI commands via setuptools entry-points.)

• cligj 0.4.0 (Click params for commmand line interfaces to GeoJSON)

• clrmagic 0.0.1a2 (IPython cell magic to use .NET languages)

• cyordereddict 1.0.0 (Cython implementation of Python's collections.OrderedDict)

• descartes 1.1.0 (Use geometric objects as matplotlib paths and patches)

• fiona 1.7.13 (reads and writes spatial data files)

• gast 0.2.0 (Python AST that abstracts the underlying Python version)

• geopandas 0.4.0 (Geographic pandas extensions)

• grpcio 1.13.0 (HTTP/2-based RPC framework)

• mlxtend 0.13.0 (Machine Learning Library Extensions)

• munch 2.3.2 (A dot-accessible dictionary (a la JavaScript objects).)

• prometheus_client 0.3.1 (Python client for the Prometheus monitoring system.)

• protobuf 3.6.0 (Protocol Buffers - Google's data interchange format)

• pyproj 1.9.5.1 (Python interface to PROJ.4 library)

• pythonnet 2.4.0.dev0 (.Net and Mono integration for Python)

• shapely 1.6.4.post1 (Geometric objects, predicates, and operations)

• tensorboard 1.9.0 (TensorBoard lets you watch Tensors Flow)

• tensorflow 1.9.0 (TensorFlow helps the tensors flow)

• termcolor 1.1.0 (ANSII Color formatting for output in terminal)

• vega_datasets 0.5.0 (A Python package for offline access to Vega datasets)

Upgraded packages:

• astroid 1.6.5 → 2.0.4 (Rebuild a new abstract syntax tree from Python's ast (required for pylint))

• beautifulsoup4 4.6.0 → 4.6.1 (Screen-scraping library)

• bokeh 0.13.0 → 1.0.0.dev5 (Statistical and novel interactive HTML plots for Python)

• bqplot 0.10.5 → 0.11.0 (Interactive plotting for the Jupyter notebook, using d3.js and ipywidgets.)

• comtypes 1.1.4 → 1.1.7 (Pure Python COM package)

• cython 0.28.3 → 0.28.5 (Cython is a language that makes writing C extensions for the Python language as easy as Python)

• dask 0.18.1 → 0.18.2 (Minimal task scheduling abstraction)

• distributed 1.22.0 → 1.22.1 (Distributed computing)

• fast_histogram 0.4 → 0.5 (Fast 1D and 2D histogram functions in Python)

• geopy 1.14.0 → 1.16.0 (Python Geocoding Toolbox)

• greenlet 0.4.13 → 0.4.14 (Lightweight in-process concurrent programming)

• hvplot 0.2.0 → 0.2.1 (A high-level plotting API for pandas, dask, streamz and xarray built on HoloViews)

• ipympl 0.2.0 → 0.2.1 (Matplotlib Jupyter Extension)

• ipython 6.4.0 → 6.5.0 (Enhanced Python shell)

• ipywidgets 7.2.1 → 7.4.0 (IPython HTML widgets for Jupyter)

• joblib 0.12.0 → 0.12.2 (Lightweight pipelining: using Python functions as pipeline jobs.)

• jupyter_sphinx 0.1.2 → 0.1.4 (Jupyter Sphinx Extensions)

• jupyterlab 0.32.1 → 0.33.8 (Jupyter lab environment notebook server extension)

• jupyterlab_launcher 0.10.5 → 0.11.2 (Jupyter Launcher)

• keras 2.2.0 → 2.2.2 (Theano-based Deep Learning library)

• keras_applications 1.0.2 → 1.0.4 (Reference implementations of popular deep learning models)

• keras_preprocessing 1.0.1 → 1.0.2 (Easy data preprocessing and data augmentation for deep learning models)

• loky 2.1.4 → 2.2.0 (Robust and reusable Executor for joblib)

• lxml 4.2.3 → 4.2.4 (Powerful and Pythonic XML processing library combining libxml2/libxslt with the ElementTree API.)

• matplotlib 2.2.2 → 2.2.3 (2D plotting library (embeddable in GUIs created with PyQt))

• more_itertools 4.2.0 → 4.3.0 (More routines for operating on iterables, beyond itertools)

• msgpack_numpy 0.4.3 → 0.4.3.1 (Numpy data serialization using msgpack)

• multipledispatch 0.5.0 → 0.6.0 (A relatively sane approach to multiple dispatch in Python)

• mypy 0.610 → 0.620 (Optional static typing for Python)

• mysql_connector_python 8.0.6 → 8.0.11 (MySQL driver written in Python)

• netcdf4 1.4.0 → 1.4.1 (Provides an object-oriented python interface to the netCDF version 4 library)

• notebook 5.5.0 → 5.6.0 (# Jupyter Notebook)

• numexpr 2.6.5 → 2.6.7 (Fast evaluation of array expressions elementwise by using a vector-based virtual machine)

• pandas 0.23.3 → 0.23.4 (Powerful data structures for data analysis, time series and statistics)

• parambokeh 0.2.2 → 0.2.3 (Declarative Python programming using Parameters.)

• paramnb 2.0.2 → 2.0.4 (Generate ipywidgets from Parameterized objects in the notebook)

• parso 0.3.0 → 0.3.1 (A Python Parser)

• pillow 5.2.0 → 5.1.1 (Python Imaging Library (fork))

• pip 10.0.1 → 18.0 (A tool for installing and managing Python packages)

• plotnine 0.3.0 → 0.4.0 (A grammar of graphics for python)

• pluggy 0.6.0 → 0.7.1 (plugin and hook calling mechanisms for python)

• pyct 0.4.3 → 0.4.5 (python package common tasks for users (e.g. copy examples, fetch data, ...))

• pygame 1.9.3 → 1.9.4 (Pygame gives multimedia to python.)

• pylint 1.9.2 → 2.1.1 (Logilab code analysis module: analyzes Python source code looking for bugs and signs of poor quality)

• pymc3 3.4.1 → 3.5 (Markov Chain Monte Carlo sampling toolkit.)

• pymongo 3.7.0 → 3.7.1 (Python driver for MongoDB http://www.mongodb.org)

• pytest 3.6.3 → 3.7.1 (pytest: simple powerful testing with Python)

• pyzmq 17.0.0 → 17.1.2 (Lightweight and super-fast messaging based on ZeroMQ library (required for IPython Qt console))

• pyzo 4.5.2 → 4.6.0 (the Python IDE for scientific computing)

• regex 2018.6.21 → 2018.7.11 (Alternative regular expression module, to replace re.)

• reportlab 3.4.0 → 3.5.2 (The PDF generation library)

• rope 0.10.7 → 0.11.0 (a python refactoring library...)

• ruamel.yaml 0.15.42 → 0.15.52 (a YAML parser/emitter that supports roundtrip preservation of comments, seq/map flow style, and map key order)

• scikit_learn 0.19.1 → 0.19.2 (A set of Python modules for machine learning and data mining)

• scs 1.2.7 → 2.0.2 (scs: splitting conic solver)

• seaborn 0.9.dev0 → 0.9.0 (statistical data visualization)

• setuptools 39.2.0 → 40.0.0 (Download, build, install, upgrade, and uninstall Python packages - easily)

• snakeviz 0.4.2 → 1.0.0 (An in-browser Python profile viewer)

• sphinx 1.7.5 → 1.7.6 (Tool for generating documentation which uses reStructuredText as its markup language)

• sphinx_rtd_theme 0.4.0 → 0.4.1 (ReadTheDocs.org theme for Sphinx, 2013 version.)

• spyder 3.3.0 → 3.3.1.dev0 (The Scientific Python Development Environment: An IDE designed for interactive computing and data visualisation with a simple and intuitive user interface)

• sqlalchemy 1.2.9 → 1.2.10 (SQL Toolkit and Object Relational Mapper)

• sympy 1.1.1 → 1.2 (Symbolic Mathematics Library)

• tornado 5.0.2 → 5.1 (Scalable, non-blocking web server and tools (required for IPython notebook))

• tqdm 4.23.4 → 4.24.0 (A Simple Python Progress Meter)

• widgetsnbextension 3.2.1 → 3.4.0 (IPython HTML widgets for Jupyter)

• winpython 1.10.20180624 → 1.10.20180811 (WinPython distribution tools, including WPPM (package manager))

• wordcloud 1.4.1 → 1.5.0 (A little word cloud generator)

• xarray 0.10.7 → 0.10.8 (N-D labeled arrays and datasets in Python)

Removed packages:

• torch 0.4.0 (a deep learning framework.)

• torchvision 0.2.1 (Datasets, Transforms and Models specific to Computer Vision)

@stonebig ,
What about PyTorch?
Version 0.4.1 is out.

This is a truly Python based Deep Learning framework as opposed to TensorFlow.

Tensorflow and Pytorch have both some advantages, see https://medium.com/@UdacityINDIA/tensorflow-or-pytorch-the-force-is-strong-with-which-one-68226bb7dab4 , but at the moment:

• Pytorch wheel as grown larger than Tensorflow, with Tensorflow growing less
• Pymc4 team has choosen Tensorflow+tensorflow-probability, with a summer of code https://sharanry.github.io/categories/gsoc18/

So for This iteration, I'm trying back the Tensorflow way, but maybe none of them will be able to remain in the Size Limit of WinPython.

build3 2018-08-20 (Jupyterlab-0.34.0 beta4)

Changes from WinPython 2018-03 build2:

• New Packages:
  . tensorflow_probability (Python-3.6 64 bit only)
• Upgraded package;
  . Spyder-3.3.1 official, Jupyterlab-0.34.0 (beta4), nodejs-8.9.4,
  . holoviews-1.11.0a3, PyQtgraph-20180818, altair-2.2.2, bqplot-1.11.1

focus for next build:

• scikit-learn-0.20.0rc, bokeh-1.0.0 (or closer), pdvega
• may be an rc, if still no news on Python-3.7.1 release date

History of changes for WinPython-64bit 3.6.6.2Qt5 b3

The following changes were made to WinPython-64bit distribution since version 3.6.6.1Qt5.

Tools

Upgraded packages:

• Nodejs v8.11.2 → v8.11.4 (a JavaScript runtime built on Chrome's V8 JavaScript engine)

Python packages

New packages:

• absl_py 0.4.0 (Abseil Python Common Libraries)

• astor 0.7.1 (Read/rewrite/write Python ASTs)

• click_plugins 1.0.3 (An extension module for click to enable registering CLI commands via setuptools entry-points.)

• cligj 0.4.0 (Click params for commmand line interfaces to GeoJSON)

• descartes 1.1.0 (Use geometric objects as matplotlib paths and patches)

• fiona 1.7.13 (reads and writes spatial data files)

• gast 0.2.0 (Python AST that abstracts the underlying Python version)

• geopandas 0.4.0 (Geographic pandas extensions)

• grpcio 1.13.0 (HTTP/2-based RPC framework)

• munch 2.3.2 (A dot-accessible dictionary (a la JavaScript objects).)

• prometheus_client 0.3.1 (Python client for the Prometheus monitoring system.)

• protobuf 3.6.1 (Protocol Buffers - Google's data interchange format)

• pyproj 1.9.5.1 (Python interface to PROJ.4 library)

• shapely 1.6.4.post1 (Geometric objects, predicates, and operations)

• tensorboard 1.10.0 (TensorBoard lets you watch Tensors Flow)

• tensorflow 1.10.0 (TensorFlow helps the tensors flow)

• tensorflow_probability 0.3.0 ()

• termcolor 1.1.0 (ANSII Color formatting for output in terminal)

Upgraded packages:

• altair 2.1.0 → 2.2.2 (High-level declarative visualization library for Python)

• astroid 1.6.5 → 2.0.4 (Rebuild a new abstract syntax tree from Python's ast (required for pylint))

• beautifulsoup4 4.6.0 → 4.6.3 (Screen-scraping library)

• bleach 2.1.3 → 2.1.4 (An easy whitelist-based HTML-sanitizing tool)

• bokeh 0.13.0 → 1.0.0.dev5 (Statistical and novel interactive HTML plots for Python)

• bqplot 0.10.5 → 0.11.1 (Interactive plotting for the Jupyter notebook, using d3.js and ipywidgets.)

• certifi 2018.4.16 → 2018.8.13 (Python package for providing Mozilla's CA Bundle.)

• cftime 1.0.0 → 1.0.1 (time-handling functionality from netcdf4-python)

• comtypes 1.1.4 → 1.1.7 (Pure Python COM package)

• cython 0.28.3 → 0.28.5 (Cython is a language that makes writing C extensions for the Python language as easy as Python)

• dask 0.18.1 → 0.18.2 (Minimal task scheduling abstraction)

• distributed 1.22.0 → 1.22.1 (Distributed computing)

• fast_histogram 0.4 → 0.5 (Fast 1D and 2D histogram functions in Python)

• geopy 1.14.0 → 1.16.0 (Python Geocoding Toolbox)

• greenlet 0.4.13 → 0.4.14 (Lightweight in-process concurrent programming)

• holoviews 1.11.0a1 → 1.11.0a3 (Composable, declarative data structures for building complex visualizations easily.)

• hvplot 0.2.0 → 0.2.1 (A high-level plotting API for pandas, dask, streamz and xarray built on HoloViews)

• ipympl 0.2.0 → 0.2.1 (Matplotlib Jupyter Extension)

• ipython 6.4.0 → 6.5.0 (Enhanced Python shell)

• ipywidgets 7.2.1 → 7.4.0 (IPython HTML widgets for Jupyter)

• joblib 0.12.0 → 0.12.2 (Lightweight pipelining: using Python functions as pipeline jobs.)

• jupyter_sphinx 0.1.2 → 0.1.4 (Jupyter Sphinx Extensions)

• jupyterlab 0.32.1 → 0.34.0 (Jupyter lab environment notebook server extension)

• jupyterlab_launcher 0.10.5 → 0.13.1 (Jupyter Launcher)

• keras 2.2.0 → 2.2.2 (Theano-based Deep Learning library)

• keras_applications 1.0.2 → 1.0.4 (Reference implementations of popular deep learning models)

• keras_preprocessing 1.0.1 → 1.0.2 (Easy data preprocessing and data augmentation for deep learning models)

• loky 2.1.4 → 2.2.0 (Robust and reusable Executor for joblib)

• lxml 4.2.3 → 4.2.4 (Powerful and Pythonic XML processing library combining libxml2/libxslt with the ElementTree API.)

• matplotlib 2.2.2 → 2.2.3 (2D plotting library (embeddable in GUIs created with PyQt))

• more_itertools 4.2.0 → 4.3.0 (More routines for operating on iterables, beyond itertools)

• msgpack_numpy 0.4.3 → 0.4.3.1 (Numpy data serialization using msgpack)

• multipledispatch 0.5.0 → 0.6.0 (A relatively sane approach to multiple dispatch in Python)

• mypy 0.610 → 0.620 (Optional static typing for Python)

• mysql_connector_python 8.0.6 → 8.0.11 (MySQL driver written in Python)

• netcdf4 1.4.0 → 1.4.1 (Provides an object-oriented python interface to the netCDF version 4 library)

• notebook 5.5.0 → 5.6.0 (# Jupyter Notebook)

• numexpr 2.6.5 → 2.6.8 (Fast evaluation of array expressions elementwise by using a vector-based virtual machine)

• pandas 0.23.3 → 0.23.4 (Powerful data structures for data analysis, time series and statistics)

• parambokeh 0.2.2 → 0.2.3 (Declarative Python programming using Parameters.)

• paramnb 2.0.2 → 2.0.4 (Generate ipywidgets from Parameterized objects in the notebook)

• parso 0.3.0 → 0.3.1 (A Python Parser)

• pillow 5.2.0 → 5.1.1 (Python Imaging Library (fork))

• pip 10.0.1 → 18.0 (A tool for installing and managing Python packages)

• plotnine 0.3.0 → 0.4.0 (A grammar of graphics for python)

• pluggy 0.6.0 → 0.7.1 (plugin and hook calling mechanisms for python)

• psutil 5.4.6 → 5.4.7 (Provides an interface for retrieving information on all running processes and system utilization (CPU, disk, memory, network) in a portable way)

• pyct 0.4.3 → 0.4.5 (python package common tasks for users (e.g. copy examples, fetch data, ...))

• pygame 1.9.3 → 1.9.4 (Pygame gives multimedia to python.)

• pylint 1.9.2 → 2.1.1 (Logilab code analysis module: analyzes Python source code looking for bugs and signs of poor quality)

• pymc3 3.4.1 → 3.5 (Markov Chain Monte Carlo sampling toolkit.)

• pymongo 3.7.0 → 3.7.1 (Python driver for MongoDB http://www.mongodb.org)

• pyodbc 4.0.23 → 4.0.24 (DB API Module for ODBC)

• pytest 3.6.3 → 3.7.1 (pytest: simple powerful testing with Python)

• pywin32_ctypes 0.1.2 → 0.1.3 (A (partial) reimplementation of pywin32 that is pure python (uses ctypes/cffi))

• pyzmq 17.0.0 → 17.1.2 (Lightweight and super-fast messaging based on ZeroMQ library (required for IPython Qt console))

• pyzo 4.5.2 → 4.6.0 (the Python IDE for scientific computing)

• qtconsole 4.3.1 → 4.4.0 (Jupyter Qt console)

• regex 2018.6.21 → 2018.7.11 (Alternative regular expression module, to replace re.)

• reportlab 3.4.0 → 3.5.5 (The PDF generation library)

• rope 0.10.7 → 0.11.0 (a python refactoring library...)

• ruamel.yaml 0.15.42 → 0.15.52 (a YAML parser/emitter that supports roundtrip preservation of comments, seq/map flow style, and map key order)

• scikit_learn 0.19.1 → 0.19.2 (A set of Python modules for machine learning and data mining)

• scs 1.2.7 → 2.0.2 (scs: splitting conic solver)

• seaborn 0.9.dev0 → 0.9.0 (statistical data visualization)

• setuptools 39.2.0 → 40.1.0 (Download, build, install, upgrade, and uninstall Python packages - easily)

• snakeviz 0.4.2 → 1.0.0 (An in-browser Python profile viewer)

• sphinx 1.7.5 → 1.7.6 (Tool for generating documentation which uses reStructuredText as its markup language)

• sphinx_rtd_theme 0.4.0 → 0.4.1 (ReadTheDocs.org theme for Sphinx, 2013 version.)

• spyder 3.3.0 → 3.3.1 (The Scientific Python Development Environment: An IDE designed for interactive computing and data visualisation with a simple and intuitive user interface)

• spyder_kernels 0.2.4 → 0.2.6 (Jupyter kernels for the Spyder console)

• sqlalchemy 1.2.9 → 1.2.10 (SQL Toolkit and Object Relational Mapper)

• sympy 1.1.1 → 1.2 (Symbolic Mathematics Library)

• tornado 5.0.2 → 5.1 (Scalable, non-blocking web server and tools (required for IPython notebook))

• tqdm 4.23.4 → 4.24.0 (A Simple Python Progress Meter)

• widgetsnbextension 3.2.1 → 3.4.0 (IPython HTML widgets for Jupyter)

• winpython 1.10.20180624 → 1.10.20180811 (WinPython distribution tools, including WPPM (package manager))

• wordcloud 1.4.1 → 1.5.0 (A little word cloud generator)

• xarray 0.10.7 → 0.10.8 (N-D labeled arrays and datasets in Python)

• xlsxwriter 1.0.5 → 1.0.6 (A Python module for creating Excel XLSX files.)

Removed packages:

• torch 0.4.0 (a deep learning framework.)

• torchvision 0.2.1 (Datasets, Transforms and Models specific to Computer Vision)

@RoyiAvital from Pybay you can get the last Travis Oliphant talk "State of AI/ML in Python": https://speakerdeck.com/teoliphant/ml-in-python

We shall hope a nice convergence of currently conflicting framework within 2 years time, as nobody wants to be trapped in a specific and incompatible framework.

@stonebig ,
Thank you for sharing.

By the way, I'd prefer Chainer + CuPy instead of TensorFlow.
Because you should, in my opinion, promote solutions based on Python and not just wrappers.

@RoyiAvital Cupy is currently 100% tied to cuda/nvidia: 0% of PCs you can find in schools.

But Chainer isn't.
People will use Chainer.

In case you have nVidia, Chainer will have the same features as PyTorch.
CuPy is just the CUDA framework of Chainer.

Just like you don't say - The GPU framework of TensorFlow / PyTorch is 100% CUDA hence 0% support on schools and remove it it is OK to include CuPy with Chainer.

Hi, By the way, could you add RISE to the packages?
It would be great to have it working out of the box.

ok, will remove pymc3 to make room.

@stonebig , Are you referring to the RISE request or PyTorch / Chainer?

Update
I see RISE is already added few days ago.
Great news 👍 .

rc (2018-08-27)

Change from WinPython 2018-03 build3

• new package: rise-5.4.1, ppci-0.5.6
• upgrade: jupyterlab-0.34.4,
• removal: Theano, pymc3

History of changes for WinPython-64bit 3.6.6.2Qt5 rc

The following changes were made to WinPython-64bit distribution since version 3.6.6.1Qt5.

Tools

Upgraded packages:

• Nodejs v8.11.2 → v8.11.4 (a JavaScript runtime built on Chrome's V8 JavaScript engine)

Python packages

New packages:

• absl_py 0.4.0 (Abseil Python Common Libraries)

• astor 0.7.1 (Read/rewrite/write Python ASTs)

• click_plugins 1.0.3 (An extension module for click to enable registering CLI commands via setuptools entry-points.)

• cligj 0.4.0 (Click params for commmand line interfaces to GeoJSON)

• descartes 1.1.0 (Use geometric objects as matplotlib paths and patches)

• fiona 1.7.13 (reads and writes spatial data files)

• gast 0.2.0 (Python AST that abstracts the underlying Python version)

• geopandas 0.4.0 (Geographic pandas extensions)

• grpcio 1.13.0 (HTTP/2-based RPC framework)

• mlxtend 0.13.0 (Machine Learning Library Extensions)

• munch 2.3.2 (A dot-accessible dictionary (a la JavaScript objects).)

• ppci 0.5.6 (A compiler for ARM, X86, MSP430, xtensa and more implemented in pure Python)

• prometheus_client 0.3.1 (Python client for the Prometheus monitoring system.)

• protobuf 3.6.1 (Protocol Buffers - Google's data interchange format)

• pyproj 1.9.5.1 (Python interface to PROJ.4 library)

• pytest_runner 4.2 (Invoke py.test as distutils command with dependency resolution)

• rise 5.4.1 (Live Reveal.js Jupyter/IPython Slideshow Extension)

• shapely 1.6.4.post1 (Geometric objects, predicates, and operations)

• tensorboard 1.10.0 (TensorBoard lets you watch Tensors Flow)

• tensorflow 1.10.0 (TensorFlow helps the tensors flow)

• tensorflow_probability 0.3.0 ()

• termcolor 1.1.0 (ANSII Color formatting for output in terminal)

Upgraded packages:

• altair 2.1.0 → 2.2.2 (High-level declarative visualization library for Python)

• astroid 1.6.5 → 2.0.4 (Rebuild a new abstract syntax tree from Python's ast (required for pylint))

• beautifulsoup4 4.6.0 → 4.6.3 (Screen-scraping library)

• bleach 2.1.3 → 2.1.4 (An easy whitelist-based HTML-sanitizing tool)

• bokeh 0.13.0 → 1.0.0.dev6 (Statistical and novel interactive HTML plots for Python)

• bqplot 0.10.5 → 0.11.1 (Interactive plotting for the Jupyter notebook, using d3.js and ipywidgets.)

• certifi 2018.4.16 → 2018.8.13 (Python package for providing Mozilla's CA Bundle.)

• cftime 1.0.0 → 1.0.1 (time-handling functionality from netcdf4-python)

• cloudpickle 0.5.3 → 0.5.5 (Extended pickling support for Python objects)

• comtypes 1.1.4 → 1.1.7 (Pure Python COM package)

• cvxpy 1.0.6 → 1.0.8 (A domain-specific language for modeling convex optimization problems in Python)

• cython 0.28.3 → 0.28.5 (Cython is a language that makes writing C extensions for the Python language as easy as Python)

• dask 0.18.1 → 0.18.2 (Minimal task scheduling abstraction)

• distributed 1.22.0 → 1.22.1 (Distributed computing)

• fast_histogram 0.4 → 0.5 (Fast 1D and 2D histogram functions in Python)

• fastparquet 0.1.5 → 0.1.6 (Python support for Parquet file format)

• fuzzywuzzy 0.16.0 → 0.17.0 (Fuzzy string matching in python)

• geopy 1.14.0 → 1.16.0 (Python Geocoding Toolbox)

• greenlet 0.4.13 → 0.4.14 (Lightweight in-process concurrent programming)

• holoviews 1.11.0a1 → 1.11.0a3 (Composable, declarative data structures for building complex visualizations easily.)

• hvplot 0.2.0 → 0.2.1 (A high-level plotting API for pandas, dask, streamz and xarray built on HoloViews)

• ipympl 0.2.0 → 0.2.1 (Matplotlib Jupyter Extension)

• ipython 6.4.0 → 6.5.0 (Enhanced Python shell)

• ipywidgets 7.2.1 → 7.4.0 (IPython HTML widgets for Jupyter)

• joblib 0.12.0 → 0.12.2 (Lightweight pipelining: using Python functions as pipeline jobs.)

• jupyter_sphinx 0.1.2 → 0.1.4 (Jupyter Sphinx Extensions)

• jupyterlab 0.32.1 → 0.34.4 (Jupyter lab environment notebook server extension)

• jupyterlab_launcher 0.10.5 → 0.13.1 (Jupyter Launcher)

• keras 2.2.0 → 2.2.2 (Theano-based Deep Learning library)

• keras_applications 1.0.2 → 1.0.4 (Reference implementations of popular deep learning models)

• keras_preprocessing 1.0.1 → 1.0.2 (Easy data preprocessing and data augmentation for deep learning models)

• loky 2.1.4 → 2.2.1 (Robust and reusable Executor for joblib)

• lxml 4.2.3 → 4.2.4 (Powerful and Pythonic XML processing library combining libxml2/libxslt with the ElementTree API.)

• matplotlib 2.2.2 → 2.2.3 (2D plotting library (embeddable in GUIs created with PyQt))

• more_itertools 4.2.0 → 4.3.0 (More routines for operating on iterables, beyond itertools)

• msgpack_numpy 0.4.3 → 0.4.3.1 (Numpy data serialization using msgpack)

• multipledispatch 0.5.0 → 0.6.0 (A relatively sane approach to multiple dispatch in Python)

• mypy 0.610 → 0.620 (Optional static typing for Python)

• mysql_connector_python 8.0.6 → 8.0.11 (MySQL driver written in Python)

• netcdf4 1.4.0 → 1.4.1 (Provides an object-oriented python interface to the netCDF version 4 library)

• notebook 5.5.0 → 5.6.0 (# Jupyter Notebook)

• numexpr 2.6.5 → 2.6.8 (Fast evaluation of array expressions elementwise by using a vector-based virtual machine)

• pandas 0.23.3 → 0.23.4 (Powerful data structures for data analysis, time series and statistics)

• parambokeh 0.2.2 → 0.2.3 (Declarative Python programming using Parameters.)

• paramnb 2.0.2 → 2.0.4 (Generate ipywidgets from Parameterized objects in the notebook)

• parso 0.3.0 → 0.3.1 (A Python Parser)

• pdvega 0.1 → 0.2.1.dev0 (Pandas plotting interface to Vega and Vega-Lite)

• pip 10.0.1 → 18.0 (A tool for installing and managing Python packages)

• plotnine 0.3.0 → 0.4.0 (A grammar of graphics for python)

• pluggy 0.6.0 → 0.7.1 (plugin and hook calling mechanisms for python)

• psutil 5.4.6 → 5.4.7 (Provides an interface for retrieving information on all running processes and system utilization (CPU, disk, memory, network) in a portable way)

• pyarrow 0.9.0 → 0.10.0 (Python library for Apache Arrow)

• pyct 0.4.3 → 0.4.5 (python package common tasks for users (e.g. copy examples, fetch data, ...))

• pygame 1.9.3 → 1.9.4 (Pygame gives multimedia to python.)

• pylint 1.9.2 → 2.1.1 (Logilab code analysis module: analyzes Python source code looking for bugs and signs of poor quality)

• pymongo 3.7.0 → 3.7.1 (Python driver for MongoDB http://www.mongodb.org)

• pyodbc 4.0.23 → 4.0.24 (DB API Module for ODBC)

• pytest 3.6.3 → 3.7.2 (pytest: simple powerful testing with Python)

• pywin32_ctypes 0.1.2 → 0.2.0 (A (partial) reimplementation of pywin32 that is pure python (uses ctypes/cffi))

• pyzmq 17.0.0 → 17.1.2 (Lightweight and super-fast messaging based on ZeroMQ library (required for IPython Qt console))

• pyzo 4.5.2 → 4.6.0 (the Python IDE for scientific computing)

• qtconsole 4.3.1 → 4.4.1 (Jupyter Qt console)

• qtpy 1.4.2 → 1.5.0 (Provides an abstraction layer on top of the various Qt bindings (PyQt5, PyQt4 and PySide) and additional custom QWidgets.)

• regex 2018.6.21 → 2018.8.17 (Alternative regular expression module, to replace re.)

• reportlab 3.4.0 → 3.5.6 (The PDF generation library)

• rope 0.10.7 → 0.11.0 (a python refactoring library...)

• ruamel.yaml 0.15.42 → 0.15.52 (a YAML parser/emitter that supports roundtrip preservation of comments, seq/map flow style, and map key order)

• scikit_learn 0.19.1 → 0.19.2 (A set of Python modules for machine learning and data mining)

• scs 1.2.7 → 2.0.2 (scs: splitting conic solver)

• seaborn 0.9.dev0 → 0.9.0 (statistical data visualization)

• setuptools 39.2.0 → 40.2.0 (Download, build, install, upgrade, and uninstall Python packages - easily)

• snakeviz 0.4.2 → 1.0.0 (An in-browser Python profile viewer)

• sphinx 1.7.5 → 1.7.7 (Tool for generating documentation which uses reStructuredText as its markup language)

• sphinx_rtd_theme 0.4.0 → 0.4.1 (ReadTheDocs.org theme for Sphinx, 2013 version.)

• spyder 3.3.0 → 3.3.1 (The Scientific Python Development Environment: An IDE designed for interactive computing and data visualisation with a simple and intuitive user interface)

• spyder_kernels 0.2.4 → 0.2.6 (Jupyter kernels for the Spyder console)

• sqlalchemy 1.2.9 → 1.2.11 (SQL Toolkit and Object Relational Mapper)

• sympy 1.1.1 → 1.2 (Symbolic Mathematics Library)

• tornado 5.0.2 → 5.1 (Scalable, non-blocking web server and tools (required for IPython notebook))

• tqdm 4.23.4 → 4.25.0 (A Simple Python Progress Meter)

• vega 1.3.0 → 1.4.0 (An IPython/ Jupyter widget for Vega and Vega-Lite)

• widgetsnbextension 3.2.1 → 3.4.0 (IPython HTML widgets for Jupyter)

• winpython 1.10.20180624 → 1.10.20180827 (WinPython distribution tools, including WPPM (package manager))

• wordcloud 1.4.1 → 1.5.0 (A little word cloud generator)

• xarray 0.10.7 → 0.10.8 (N-D labeled arrays and datasets in Python)

• xlsxwriter 1.0.5 → 1.0.7 (A Python module for creating Excel XLSX files.)

• xlwings 0.11.5 → 0.11.8 (Interact with Excel from Python and vice versa)

Removed packages:

• pymc3 3.4.1 (Markov Chain Monte Carlo sampling toolkit.)

• theano 1.0.2 (Optimizing compiler for evaluating mathematical expressions on CPUs and GPUs.)

• torch 0.4.0 (a deep learning framework.)

• torchvision 0.2.1 (Datasets, Transforms and Models specific to Computer Vision)

• vega3 0.13.0 (An IPython/ Jupyter widget for Vega 3 and Vega-Lite 2)

uploading rc as final.

Congrats!

What about Chainer and CuPy?

at the moment:

• cupy works only with a nvidia gpu, so chainer+cupy is not an option for the target audience (education),
• I just see I included IDLEX only in the small zero version.... shame, have to re-spin an "rc2"

Hi,

The GPU features of TensorFlow only works on nVidia GPU's as well yet it is included.
Chianer is like PyTorch.
Just in the case of PyTorch the GPU framework is part of PyTorch while in Chainer the GPU framework is called CuPy.

If you add Chainer + CuPy you'll have Dynamic Graph DNN which is written in Python with Numpy acceleration for those with nVidia GPU. Just like TensorFlow / PyTorch or any other DNN library.

I'm coming from education (Student).
In my courses we try to use pure Python hence TensorFlow isn't good.
PyTorch / Chainer is perfect for that (Education) as everything is written in Python.

rc3 (2018-09-02) for the non-zero wheels:

change from rc1:

• include IDLEX in the bigger wheels, (IDLEX was only in the "zero" wheels),
• need to upgrade Jupyterlab/ipywidgets so it builds properly.

@stonebig ,
Could you please address my post before?
It will really assist the students in my course.

@RoyiAvital you may:

• remain on previous release for your student course, as it was apparently ok for you
• or use the next version, when ready, with additional steps :
  • download Torch (and Torchivision ?) in a dedicated place , for example M:\complementary_wheels:
    • pip download --dest M:\complementary_wheels http://download.pytorch.org/whl/cpu/torch-0.4.1-cp36-cp36m-win_amd64.whl
    • pip download --dest M:\complementary_wheels torchvision
  • install it on each student pc per:
    • pip install torch torchvision --no-index --trusted-host=None --find-links=M:\complementary_wheels

I'm strugling to make Jupyterlab build again, so you may anyway not have a new version until a few days.

What I don't get is why would you prefer a library you can't see how it works and learn form it (TensorFlow) over options which are pure Python and people can look at the code, debug it and learn from it.

Isn't it better for your audience (Education)?

@RoyiAvital the choice is made of Tensorflow for this release. We'll see For next one:

• it must fit in nullsoft installer size,
• it must have a good bayesian statistics package (found only Pyro-ppl https://github.com/uber/pyro, not sure it's nice with windows: they don't test)

I don't remember Pytorch was on Christoph Gohlke "reference" list before
today https://www.lfd.uci.edu/~gohlke/pythonlibs/#pytorch

rc3 is the final version... updating website soon