Plotting Spikes,weights,Performance about snnlibpy

[HiroshiARAKI]

Hi.

I have no idea about your problem completely.
So, before I answer your question, I have some questions for you.

1. Do you know what error message said? If you know that please tell me.
2. Do you use GPU computing??
3. Didn't all plotting function work? (e.g. "Only Plotting spikes work properly," "only plotting weights doesn't work well, " or something like that)
4. If your network is before training, all plotting functions work well??

that's all my questions.
If these questions' answers are crystal-clear, I may solve your problem.

[tehreemnaqvi]

Hi, Actually i am trying to plot these spikes after training.I am also using GPU and once training is complete, none of the spikes are generated.I didn't get any error message after training is completed.

[HiroshiARAKI]

Hmm... I've not known well about that problem yet, but...
If you put on "plt.close()" after "plt.show()" or "plt.savefig()", how does your program work?

[tehreemnaqvi]

Hi,This is my code.
Can you see what's the problem in this code?

for epoch in range( args.n_epochs ):
    # Create a dataloader to iterate over dataset.
    dataloader = DataLoader(
        dataset,
        batch_size=args.batch_size,
        shuffle=True,
        num_workers=args.n_workers,
        pin_memory=args.gpu,
    )

    for step, batch in enumerate( tqdm( dataloader ) ):
        # Prep next input batch.
        inpts = {"I": batch["encoded_image"]}
        if args.gpu:
            inpts = {k: v.cuda() for k, v in inpts.items()}

        # Run the network on the input.
        network.run( inputs=inpts, time=args.time )

        # Plot simulation data.
        if args.plot:
            spikes = {}
            for name, monitor in network.monitors.items():
                spikes[name] = monitor.get( "s" )[:, 0].view( args.time, -1 )

            spike_ims, spike_axes = plot_spikes(
                spikes, ims=spike_ims, axes=spike_axes
            )

            conv1_weights_im = plot_conv2d_weights(
                conv1_connection.w, im=conv1_weights_im, wmin=-1.0, wmax=1.0
            )
            conv2_weights_im = plot_conv2d_weights(
                conv2_connection.w, im=conv2_weights_im, wmin=-1.0, wmax=1.0
            )
            dense_weights_im = plot_weights(
                dense_connection.w, im=dense_weights_im, wmin=-1.0, wmax=1.0
            )
            output_weights_im = plot_weights(
                output_connection.w, im=output_weights_im, wmin=-1.0, wmax=1.0
            )
            #plt.ioff()
           
            plt.show()
            plt.close()

        network.reset_state_variables()

[tehreemnaqvi]

When i run the program after calling plt.close().I had the same output and did not get any plot.
This is the output after training.

[HiroshiARAKI]

Hi,
I can't find where the essential problem is.

How about that you stop giving 'im (AxesImage object)' as arguments and change to call "plt.show" or "plt.savefig()" after each plotting functions.
So, what happens if you modify a code as below.

plot_conv2d_weights(conv1_connection.w,  wmin=-1.0, wmax=1.0)
plt.savefig('****.png')
plt.close()

plot_conv2d_weights(conv2_connection.w, wmin=-1.0, wmax=1.0)
plt.savefig('****.png')
plt.close()

plot_weights(dense_connection.w, wmin=-1.0, wmax=1.0)
plt.savefig('****.png')
plt.close()

plot_weights(output_connection.w, wmin=-1.0, wmax=1.0)
plt.savefig('****.png')
plt.close()

If it does not work well, you try to do with one plotting function.
For example, only

plot_weights(output_connection.w, wmin=-1.0, wmax=1.0)
plt.savefig('****.png')
plt.close()

this code.

Maybe, this problem is about matplotlib, not bindsnet.

[tehreemnaqvi]

Thank you very much. Now, i will try this method.

…

On Mon, 11 May 2020 at 11:30, Hiroshi ARAKI ***@***.***> wrote: Hi, I can't find where the essential problem is. How about that you stop giving 'im (AxesImage object)' as arguments and change to call "plt.show" or "plt.savefig()" after each plotting functions. So, what happens if you modify a code as below. ` plot_conv2d_weights(conv1_connection.w, wmin=-1.0, wmax=1.0) plt.savefig('****.png') plt.close() plot_conv2d_weights(conv2_connection.w, wmin=-1.0, wmax=1.0) plt.savefig('****.png') plt.close() plot_weights(dense_connection.w, wmin=-1.0, wmax=1.0) plt.savefig('****.png') plt.close() plot_weights(output_connection.w, wmin=-1.0, wmax=1.0) plt.savefig('****.png') plt.close() ` If it does not work well, you try to do with one plotting function. For example, only plot_weights(output_connection.w, wmin=-1.0, wmax=1.0) plt.savefig('****.png') plt.close() this code. Maybe, this problem is about matplotlib, not bindsnet. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJFFXM7QDPJKHWUPX2ETHATRQ5PNRANCNFSM4M4ZR6IA> .

[HiroshiARAKI]

And, check my plotting code out.

https://github.com/HiroshiARAKI/snnlibpy/blob/master/wbn/snnlib.py#L870

This code using bindsnet's plotting functions works properly on my environment.

[tehreemnaqvi]

Thank you so much. I will refer this.

…

On Mon, 11 May 2020 at 11:42, Hiroshi ARAKI ***@***.***> wrote: And, check my plotting code out. https://github.com/HiroshiARAKI/snnlibpy/blob/master/wbn/snnlib.py#L870 This code using bindsnet's plotting functions works properly on my environment. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJFFXM4O2RYO3QIP4VK6XWDRQ5QYFANCNFSM4M4ZR6IA> .

[tehreemnaqvi]

Your code don't run on Windows?Right?

…

On Mon, 11 May 2020 at 11:42, Hiroshi ARAKI ***@***.***> wrote: And, check my plotting code out. https://github.com/HiroshiARAKI/snnlibpy/blob/master/wbn/snnlib.py#L870 This code using bindsnet's plotting functions works properly on my environment. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AJFFXM4O2RYO3QIP4VK6XWDRQ5QYFANCNFSM4M4ZR6IA> .

[tehreemnaqvi]

I tried the way you told me about plotting but again did not get any plot. On Mon, 11 May 2020 at 12:41, Tehreem Naqvi <[email protected]> wrote:

…

Your code don't run on Windows?Right? On Mon, 11 May 2020 at 11:42, Hiroshi ARAKI ***@***.***> wrote: > And, check my plotting code out. > > https://github.com/HiroshiARAKI/snnlibpy/blob/master/wbn/snnlib.py#L870 > > This code using bindsnet's plotting functions works properly on my > environment. > > — > You are receiving this because you authored the thread. > Reply to this email directly, view it on GitHub > <#1 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AJFFXM4O2RYO3QIP4VK6XWDRQ5QYFANCNFSM4M4ZR6IA> > . >

[tehreemnaqvi]

Now, this is my code: for epoch in range( args.n_epochs ): # Create a dataloader to iterate over dataset. dataloader = DataLoader( dataset, batch_size=args.batch_size, shuffle=True, num_workers=args.n_workers, pin_memory=args.gpu, ) for step, batch in enumerate( tqdm( dataloader ) ): # Prep next input batch. inpts = {"I": batch["encoded_image"]} if args.gpu: inpts = {k: v.cuda() for k, v in inpts.items()} # Run the network on the input. network.run( inputs=inpts, time=args.time ) # Plot simulation data. if args.plot: spikes = {} for name, monitor in network.monitors.items(): spikes[name] = monitor.get( "s" )[:, 0].view( args.time, -1 ) spike_ims, spike_axes = plot_spikes( spikes, ims=spike_ims, axes=spike_axes ) #conv1_weights_im = plot_conv2d_weights( #conv1_connection.w, im=conv1_weights_im, wmin=-1.0, wmax=1.0 #) plot_conv2d_weights( conv1_connection.w, wmin=-1.0, wmax=1.0 ) plt.savefig( '1.png' ) plt.close() plot_conv2d_weights( conv2_connection.w, wmin=-1.0, wmax=1.0 ) plt.savefig( '2.png' ) plt.close() plot_weights( dense_connection.w, wmin=-1.0, wmax=1.0 ) plt.savefig( '3.png' ) plt.close() plot_weights( output_connection.w, wmin=-1.0, wmax=1.0 ) plt.savefig( '4.png' ) plt.show() plt.close() #conv2_weights_im = plot_conv2d_weights( #conv2_connection.w, im=conv2_weights_im, wmin=-1.0, wmax=1.0 #) #dense_weights_im = plot_weights( #dense_connection.w, im=dense_weights_im, wmin=-1.0, wmax=1.0 #) # output_weights_im = plot_weights( #output_connection.w, im=output_weights_im, wmin=-1.0, wmax=1.0 #) #plt.ioff() #plt.draw() #plt.pause(0.001) #plt.close() network.reset_state_variables() On Mon, 11 May 2020 at 13:09, Tehreem Naqvi <[email protected]> wrote:

…

I tried the way you told me about plotting but again did not get any plot. On Mon, 11 May 2020 at 12:41, Tehreem Naqvi ***@***.***> wrote: > Your code don't run on Windows?Right? > > On Mon, 11 May 2020 at 11:42, Hiroshi ARAKI ***@***.***> > wrote: > >> And, check my plotting code out. >> >> https://github.com/HiroshiARAKI/snnlibpy/blob/master/wbn/snnlib.py#L870 >> >> This code using bindsnet's plotting functions works properly on my >> environment. >> >> — >> You are receiving this because you authored the thread. >> Reply to this email directly, view it on GitHub >> <#1 (comment)>, >> or unsubscribe >> <https://github.com/notifications/unsubscribe-auth/AJFFXM4O2RYO3QIP4VK6XWDRQ5QYFANCNFSM4M4ZR6IA> >> . >> >

[tehreemnaqvi]

Thank you very much for your help. Problem solved by plt.savefig(). On Mon, 11 May 2020 at 13:37, Tehreem Naqvi <[email protected]> wrote:

…

Now, this is my code: for epoch in range( args.n_epochs ): # Create a dataloader to iterate over dataset. dataloader = DataLoader( dataset, batch_size=args.batch_size, shuffle=True, num_workers=args.n_workers, pin_memory=args.gpu, ) for step, batch in enumerate( tqdm( dataloader ) ): # Prep next input batch. inpts = {"I": batch["encoded_image"]} if args.gpu: inpts = {k: v.cuda() for k, v in inpts.items()} # Run the network on the input. network.run( inputs=inpts, time=args.time ) # Plot simulation data. if args.plot: spikes = {} for name, monitor in network.monitors.items(): spikes[name] = monitor.get( "s" )[:, 0].view( args.time, -1 ) spike_ims, spike_axes = plot_spikes( spikes, ims=spike_ims, axes=spike_axes ) #conv1_weights_im = plot_conv2d_weights( #conv1_connection.w, im=conv1_weights_im, wmin=-1.0, wmax=1.0 #) plot_conv2d_weights( conv1_connection.w, wmin=-1.0, wmax=1.0 ) plt.savefig( '1.png' ) plt.close() plot_conv2d_weights( conv2_connection.w, wmin=-1.0, wmax=1.0 ) plt.savefig( '2.png' ) plt.close() plot_weights( dense_connection.w, wmin=-1.0, wmax=1.0 ) plt.savefig( '3.png' ) plt.close() plot_weights( output_connection.w, wmin=-1.0, wmax=1.0 ) plt.savefig( '4.png' ) plt.show() plt.close() #conv2_weights_im = plot_conv2d_weights( #conv2_connection.w, im=conv2_weights_im, wmin=-1.0, wmax=1.0 #) #dense_weights_im = plot_weights( #dense_connection.w, im=dense_weights_im, wmin=-1.0, wmax=1.0 #) # output_weights_im = plot_weights( #output_connection.w, im=output_weights_im, wmin=-1.0, wmax=1.0 #) #plt.ioff() #plt.draw() #plt.pause(0.001) #plt.close() network.reset_state_variables() On Mon, 11 May 2020 at 13:09, Tehreem Naqvi ***@***.***> wrote: > I tried the way you told me about plotting but again did not get any plot. > > > On Mon, 11 May 2020 at 12:41, Tehreem Naqvi ***@***.***> > wrote: > >> Your code don't run on Windows?Right? >> >> On Mon, 11 May 2020 at 11:42, Hiroshi ARAKI ***@***.***> >> wrote: >> >>> And, check my plotting code out. >>> >>> https://github.com/HiroshiARAKI/snnlibpy/blob/master/wbn/snnlib.py#L870 >>> >>> This code using bindsnet's plotting functions works properly on my >>> environment. >>> >>> — >>> You are receiving this because you authored the thread. >>> Reply to this email directly, view it on GitHub >>> <#1 (comment)>, >>> or unsubscribe >>> <https://github.com/notifications/unsubscribe-auth/AJFFXM4O2RYO3QIP4VK6XWDRQ5QYFANCNFSM4M4ZR6IA> >>> . >>> >>

[HiroshiARAKI]

Hi, you're welcome. Im glad to solve your problem.