search

ContinuumIO / anaconda-issues

Anaconda issue tracking
648 stars 222 forks source link

Navigator does not update packages #12445

Open yazabaza opened 3 years ago

yazabaza commented 3 years ago

Issue 9087 was marked as Closed but still exists. It is 2 years old and users are still reporting the same problem.

https://github.com/ContinuumIO/anaconda-issues/issues/9087

robert99 commented 3 years ago

Navigator marks a series of packages as updateable. The update runs but does not update the package. What is not clear is whether that means there is a new package but your particular system cannot update to it because of older supporting packages. Anaconda does not give any warning that this may be the case. So a user is expecting it to update but it never does. I have also had cases when a package which would update has downloaded and updated other packages that I did not expect. Very difficult to know what is going on. Sometimes conda (from Navigator) can update packages, sometime it cannot.

BabakAmini commented 3 years ago

I came here from that closed issue and have the same issue. I tried all the solutions I found on the internet but I am still unable to update some of the packages.

yazabaza commented 3 years ago

I just submitted the same issue again. Third time in 2 years. I think it's a lost cause. I gave up on Navigator long ago and just conda on command line. Anaconda Navigator is a POS.

robert99 commented 3 years ago

I did a complete clean uninstall (delete all files related to conda and spyder etc) and reinstalled the November version. I then used conda to update everything. Navigator is now on version 4 and everything seems to work now.

There is also a recent new Anaconda version you might try.

macsunmood commented 2 years ago

What if updating via conda through terminal doesn't help as well? Should I exhaustively switch to pip for this case?

robert99 commented 2 years ago

pip can work or it might clash with conda versions. Anaconda is so complex now that the chances of things not being compatible gets higher and higher. Although Navigator keeps telling you there are upgrades - when you try to install them they just effectively say the latest one that will work with your set up is already installed. I think users will have to accept the version they start with as Anaconda is not reliably updateable. It may be that you should ignore the install advice and use the set paths option. It may be in Windows that there is a space in your user name - this fault was cured but perhaps they forgot that. It may be that you have to switch off your antivirus software during installs.

Last week there was a new Anaconda version. I suggest you clean uninstall and try that - most things are up to date. Otherwise you will be spending a lot of time getting nowhere.

Here are some of my notes on clean uninstalling on my system:

image

macsunmood commented 2 years ago

@robert99 Thanks for your detailed answer. My version of Anaconda was also the latest. Uninstalled it and reinstalled it again pure fresh. Can't update/install almost nothing as of packages. What does one do when conda does this for hours with zero outcome? image

I mean, I installed fresh Anaconda, how come am I not supposed to have even tensorflow ?

robert99 commented 2 years ago

Yes, your problems are very similar to what I have been getting for the last two years. Frozen solve and flexible solve find dozens of already installed packages that are not compatible with your intended upgrade so it fails. I find it hard to believe that so many unrelated packages could be incompatible. Anaconda do not seem interested in solving it, nor even explaining what is going on. If you are using pip to install a program that has not been installed in the Anaconda package then this still works for me. Otherwise, users have to accept that Anaconda is no longer updateable. Just install it, and if it works, leave it alone.

What you could try is WinPython https://winpython.github.io/ It has a large range of the latest packages and you can use pip to update anything. Any missing programs, such as Tensorflow or PyTorch, you need can often be found ready compiled at https://www.lfd.uci.edu/~gohlke/pythonlibs/ These are also easy to install. WinPython also has the advantage that it is run as is and unbreakable - i.e. not installed to Windows - so is transportable and runs every time from the installed directory. The community is very helpful.

macsunmood commented 2 years ago

@robert99 thanks again! Would you recommend WinPython over Anaconda specifically for DS/ML purpose? Looks very interesting, but at the same time lack of official support from Tensorflow / PyTorch and such makes it dubious.

robert99 commented 2 years ago

There are about 24,000 posts on TensorFlow in their Community - that is some support! https://github.com/issues?q=tensorflow

Designed for scientists, data-scientists, and education (thanks to NumPy, SciPy, Sympy, Matplotlib, Pandas, pyqtgraph, etc.): interactive data processing and visualization using Python with Spyder and Jupyter/IPython, Pyzo, IDLEX or IDLE fully integrated Cython and Numba! See included example connectors (cffi, odbc, rpy2, scilab2py, requests, ...) for advanced users. So it is ideal for DS/ML

Anaconda Navigator has an excellent interface and I use it when it works. When it fails I use WinPython.

There are alternatives to try: ActivePython, "CPython", Enthought Canopy (new system now). Active Python has Tensorflow etc https://www.activestate.com/products/python/python-machine-learning/