agnezio/agnez

Analytics for Deep Learning projects.

agnez's Issues

Interpretation of fourth layer preferences

First of all thanks for putting your efforts on making deep visualizations available to a broader audience.
Due to the lack of other options I would like to ask you here about your DeepPref class.

If my memories are correct, visualizing normalized weights from the first layer corresponds to a close-form solution of the optimization problem 'which input maximizes 1st layer activations'. I see that something similar makes sort of sense between the 3rd and 1st layers too - as it is done in the pref_grid function. However, I fail to see any interpretation with further layers. For instance DeepPref between 4th and 1st layers returns K weights between the two first layers, where K is sorted according to the the activations of the 4th layer. What does it mean?

not able to install

Both installation methods, pip and easy_install fails. Are there other ways to install?

Thanks,
Satya

MNIST.ipynb example breaks after first epoch

Hi Eder, amazing project, really like where you are going with this. Thanks for putting in the work.

When I run the example provided in MNIST.ipynb, following the installation instructions as provided, on OSX 10.10.4, I get the following error after the first epoch:

 /my/path/agnez_app/node_modules/feathers/node_modules/feathers-commons/lib/sockets/helpers.js:61
            socket[info.method](eventName, dispatchData);
                               ^
TypeError: undefined is not a function
  at /my/path/agnez_app/node_modules/feathers/node_modules/feathers-commons/lib/sockets/helpers.js:61:32
  at Object.defaultDispatcher (/my/path/agnez_app/node_modules/feathers/node_modules/feathers-commons/lib/sockets/helpers.js:39:3)
  at /my/path/agnez_app/node_modules/feathers/node_modules/feathers-commons/lib/sockets/helpers.js:56:20
  at Array.forEach (native)
  at Object.<anonymous> (/my/path/agnez_app/node_modules/feathers/node_modules/feathers-commons/lib/sockets/helpers.js:55:35)
  at emitTwo (events.js:87:13)
  at Object.emit (events.js:169:7)
  at /my/path/agnez_app/node_modules/feathers/lib/mixins/event.js:49:21
  at Array.forEach (native)
  at EventEmitter.<anonymous> (/my/path/agnez_app/node_modules/feathers/lib/mixins/event.js:48:16)
  at emitMany (events.js:108:13)
  at EventEmitter.emit (events.js:179:7)
  at Object.exports.emitEvents (/my/path/agnez_app/node_modules/feathers/node_modules/rubberduck/lib/utils.js:32:18)
  at callbackWrapper (/my/path/agnez_app/node_modules/feathers/node_modules/rubberduck/lib/rubberduck.js:56:15)
  at /my/path/agnez_app/node_modules/feathers-nedb/lib/index.js:137:5
  at callback (/my/path/agnez_app/node_modules/feathers-nedb/node_modules/nedb/lib/executor.js:30:17)
  at Cursor.execFn (/my/path/agnez_app/node_modules/feathers-nedb/node_modules/nedb/lib/datastore.js:462:14)
  at Cursor._exec (/my/path/agnez_app/node_modules/feathers-nedb/node_modules/nedb/lib/cursor.js:172:17)
  at /my/path/agnez_app/node_modules/feathers-nedb/node_modules/nedb/lib/executor.js:40:13
  at Object.q.process (/my/path/agnez_app/node_modules/feathers-nedb/node_modules/nedb/node_modules/async/lib/async.js:731:21)
  at next (/my/path/agnez_app/node_modules/feathers-nedb/node_modules/nedb/node_modules/async/lib/async.js:728:27)
  at Immediate._onImmediate (/my/path/agnez_app/node_modules/feathers-nedb/node_modules/nedb/node_modules/async/lib/async.js:24:16)
  at processImmediate [as _immediateCallback] (timers.js:368:17)

Any idea how this comes about? Thanks.

AttributeError: 'LossPlot' object has no attribute 'totals'

When running the IPython notebook example, I get the following error:

AttributeError                            Traceback (most recent call last)
<ipython-input-6-9386fd8b8064> in <module>()
      1 model.fit(X_train, Y_train, batch_size=batch_size, nb_epoch=nb_epoch,
      2           show_accuracy=True, verbose=1, validation_data=(X_test, Y_test),
----> 3           callbacks=[loss_plot, conv_weights])

/usr/local/lib/python2.7/dist-packages/keras/models.pyc in fit(self, x, y, batch_size, nb_epoch, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, **kwargs)
    406                               shuffle=shuffle,
    407                               class_weight=class_weight,
--> 408                               sample_weight=sample_weight)
    409 
    410     def evaluate(self, x, y, batch_size=32, verbose=1,

/usr/local/lib/python2.7/dist-packages/keras/engine/training.pyc in fit(self, x, y, batch_size, nb_epoch, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight)
   1051                               verbose=verbose, callbacks=callbacks,
   1052                               val_f=val_f, val_ins=val_ins, shuffle=shuffle,
-> 1053                               callback_metrics=callback_metrics)
   1054 
   1055     def evaluate(self, x, y, batch_size=32, verbose=1, sample_weight=None):

/usr/local/lib/python2.7/dist-packages/keras/engine/training.pyc in _fit_loop(self, f, ins, out_labels, batch_size, nb_epoch, verbose, callbacks, val_f, val_ins, shuffle, callback_metrics)
    810                         for l, o in zip(out_labels, val_outs):
    811                             epoch_logs['val_' + l] = o
--> 812             callbacks.on_epoch_end(epoch, epoch_logs)
    813             if callback_model.stop_training:
    814                 break

/usr/local/lib/python2.7/dist-packages/keras/callbacks.pyc in on_epoch_end(self, epoch, logs)
     37     def on_epoch_end(self, epoch, logs={}):
     38         for callback in self.callbacks:
---> 39             callback.on_epoch_end(epoch, logs)
     40 
     41     def on_batch_begin(self, batch, logs={}):

/usr/local/lib/python2.7/dist-packages/agnez-0.1.0-py2.7.egg/agnez/app_callbacks.pyc in on_epoch_end(self, epoch, logs)
     75 
     76     def on_epoch_end(self, epoch, logs={}):
---> 77         self.train_values.append(self.totals['loss']/self.seen)
     78         self.valid_values.append(logs['val_loss'])
     79 

AttributeError: 'LossPlot' object has no attribute 'totals'

...and I don't entirely know how to fix this. Any help appreciated.

I checked and there is a self.totals in agnez/keras_callbacks.py (in the Plot class), which doesn't seem to be used here, and in keras/callbacks.py (BaseLogger class, but also not sure if used here).

My Keras is version 1.0.6, Theano version 0.9.0dev1.dev-a668c6c5b6d055b233aa5bc50b22800d996ffce1 on Python 2.7, Ubuntu 15.10 64bit.