branch: models_igr: graph display + record selection #6

igorpodolak · 2020-03-23T10:57:32Z

in utils/network_status_graph.py better display of graph: graphs are vertically centered
CORRECTED all comments, still some very minor problems arise with single edges drawn incorrectly

UPDATES to record.py

corrected all previous what was written in previous comments
CHANGED all simulationa parameters which are constant in _plot_animate() to a namedtuple

…red); marking of true/predicted classes (record.py), randomization of starting weights (ebner_agent.py)

…modele

…utput in record.py"

ziemowit-s · 2020-03-23T14:27:58Z

neuronpp/utils/record.py


                # update data
                line.set_data(t, r)
-
+                # info draw triangles for true and predicted classes


Should be parametrize from top -> plot() function, to set or no this drawing (as optional)

I shall an an option

ziemowit-s · 2020-03-23T14:29:26Z

neuronpp/utils/record.py


                # update data
                line.set_data(t, r)
-
+                # info draw triangles for true and predicted classes
+                true_x, true_y, pred_x, pred_y = self._class_tcks(label=i, true_class=true_class, pred_class=pred_class,


i is not a label. It is an enumerator of sections to records with the same variable name (eg. v)

Yes, this is an error here, or rather it is used in assumption that labels are 0, 1, 2, ... They need not to be. I shall pass a table of class_names as parameter, consistent with enumeration, and then see line in line 177 we shall have
if true_class[k] == class_names[label], and similarly in 179: if pred_class == true_classes[label]

i represents number of segment saved to be recorded. If you want to record from 5 neurons - 2 input, 3 output in the same Record object - you can. So in this case i won't represent any true values, but rather enumerator.

I think you should add parameter to the Record object to show predicted/true values, but it must be done explicitely as optional param: in the constructor or in the plotting method. Because otherwise it is extremelly error prone.

ziemowit-s · 2020-03-23T14:35:47Z

neuronpp/utils/record.py

            fig.canvas.draw()
            fig.canvas.flush_events()

        if create_fig:
            plt.show(block=False)

+    def _class_tcks(self, label, true_class, pred_class, t, r, stepsize, dt):


write full name of the function

ok, how about

Suggested change

def _class_tcks(self, label, true_class, pred_class, t, r, stepsize, dt):

def _draw_true_predicted_class_marks(self, label, true_class, pred_class, t, r, stepsize, dt):

it works for me

ziemowit-s · 2020-03-23T14:39:45Z

neuronpp/utils/record.py

                ax.set_title("Variable: %s" % var_name)
                ax.plot(self.t, rec, label=name)
                ax.set(xlabel='t (ms)', ylabel=var_name)
                ax.legend()

-    def _plot_animate(self, steps=10000, y_lim=None, position=None):
+    def _plot_animate(self, steps=10000, y_lim=None, position=None, true_class=None, pred_class=None, stepsize=None,
+                      dt=None):


If you add new params to the method with docstring - please add docstring as well for them

neuronpp/utils/record.py

wsredniawa · 2020-03-23T14:57:27Z

neuronpp/utils/network_status_graph.py


            if 'inh' in c.name:
                self.colors.append('red')
-                y_pos -= 5
+                # todo center vertically by half width of the hid layer
+                y_pos -= 6


This was original idea how to present plot but now I think it was bad idea to manually shift it like this. But it maybe as it is

Problem w linii 95 z rzeczywistym przesunięciem polega na znalezieniu nazwy warstwy poprzedzającej. Jeśli jest nią 'hid_1' to wtedy może być przez

Suggested change

y_pos -= 6

y_pos -= self.population_sizes['hid_1'] // 2 + self.population_sizes[pop_name]

igorpodolak · 2020-03-23T17:16:24Z

neuronpp/utils/record.py

+        true_y = [min(r) + 2] * len(true_x)
+        pred_y = [max(r) - 2] * len(pred_x)


Need to find better true_y (position of lower triangle) and red_y (position of upper one). All goes well as long as the motor neurone are activated.

ziemowit-s · 2020-03-24T16:40:19Z

neuronpp/utils/record.py

            fig.canvas.draw()
            fig.canvas.flush_events()

        if create_fig:
            plt.show(block=False)

+    def _class_tcks(self, label, true_class, pred_class, t, r, stepsize, dt):


it works for me

ziemowit-s · 2020-03-24T16:53:56Z

neuronpp/utils/record.py


                # update data
                line.set_data(t, r)
-
+                # info draw triangles for true and predicted classes
+                true_x, true_y, pred_x, pred_y = self._class_tcks(label=i, true_class=true_class, pred_class=pred_class,


i represents number of segment saved to be recorded. If you want to record from 5 neurons - 2 input, 3 output in the same Record object - you can. So in this case i won't represent any true values, but rather enumerator.

I think you should add parameter to the Record object to show predicted/true values, but it must be done explicitely as optional param: in the constructor or in the plotting method. Because otherwise it is extremelly error prone.

neuronpp/utils/record.py

…class labels parameter

# Conflicts: # agents/agent.py # agents/ebner_agent.py

ziemowit-s · 2020-03-26T10:25:19Z

neuronpp/utils/record.py

+                    if true_labels is not None:
+                        true_x, pred_x = self._true_predicted_class_marks(label=true_labels[i], true_class=true_class,
+                                                                          pred_class=pred_class, t=t, r=r,
+                                                                          stepsize=stepsize, dt=dt)


There is no r param in the method self._true_predicted_class_marks()

ziemowit-s · 2020-03-26T10:30:36Z

neuronpp/utils/record.py

+                        raise ValueError("True_labels parameter need to be given if show_true_prediction is True")
+                    if y_lim is None:
+                        true_y = [y_limits[0] + np.abs(y_limits[0]) * 0.09] * len(true_x)
+                        pred_y = [y_limits[1] - np.abs(y_limits[1] * 0.11)] * len(pred_x)


variables y_lim and y_limits are too confusing. It should be a single variable. I think you should override y_lim in line 154.

Actually no, because the parameter y_lim is global for all plots (a tuple in parameters); while y_limits are compoted per individual plots. During the course of simulation, the activations tend to have different values.
Changed the name y_limits to this_plot_y_limits.

Actually this shall never work as it was before and is now: if the max--min values come too near, the y_lim plots is much different from what we can read from data. There must be a method which shall first compute the wanted limits, then set them, then read what matplotlib really used. Is there one?

ziemowit-s · 2020-03-26T10:36:44Z

neuronpp/utils/record.py

+        :param pred_class: list of predicted class labels in window
+        :param stepsize: agent readout time step
+        :param dt: agent integration time step
+        :param show_true_predicted: whther to print true/predicted class' marks on the plot


I think there is no point of using show_true_predicted param, because you can just check whether true_class and pred_class is not None.

This is a matter of liking. Lots of code is written so that switch some behaviour on/off using a single switch.
But this is an important comment: I shall move out the binary show_true_predicted out of the run_params tuple. Even so as this is not actually a simulation parameter.

ziemowit-s · 2020-03-26T10:38:26Z

neuronpp/utils/record.py


                # update data
                line.set_data(t, r)
-
+                if show_true_predicted:
+                    # info draw triangles for true and predicted classes


This whole part should be moved to the separated method. The method _plot_animate() is already too big.

THe owner is my god

neuronpp/utils/record.py

ziemowit-s · 2020-03-26T10:52:21Z

neuronpp/utils/record.py

@@ -1,11 +1,9 @@
-from nrn import Segment, Section


I think this concept of recording with labels should be moved to separated class which inherits from Record, eg. RecordWithLabels. In this case - dt and true_labels params from the _plot_animate() metod can be moved into consructor.

…tuple

# Conflicts: # neuronpp/utils/record.py

igorpodolak added 4 commits March 20, 2020 18:36

Updated drawing of network (network_status_graph.py; vertically cente…

6fe2e9a

…red); marking of true/predicted classes (record.py), randomization of starting weights (ebner_agent.py)

"selection of output in record.py"

d5df424

Merge branch 'master' of https://github.com/ziemowit-s/neuronpp into …

2ddc41c

…modele

"centered display of graph in network_status_graph.py; selection of o…

e5fe31f

…utput in record.py"

ziemowit-s requested changes Mar 23, 2020

View reviewed changes

wsredniawa reviewed Mar 23, 2020

View reviewed changes

igorpodolak commented Mar 23, 2020

View reviewed changes

ziemowit-s requested changes Mar 24, 2020

View reviewed changes

igorpodolak added 4 commits March 24, 2020 18:18

Display of true/predicted classes in record.py; true_labels added as …

4fb520f

…class labels parameter

"class labels passed to record:plot() --> plot_animate()"

7758ed8

Merge remote-tracking branch 'origin/master' into models_igr

06fc25d

# Conflicts: # agents/agent.py # agents/ebner_agent.py

Plotting of outputs: record.py: true_labels added; names changed

3ed9930

ziemowit-s requested changes Mar 26, 2020

View reviewed changes

igorpodolak and others added 11 commits March 27, 2020 10:13

record.py: namedtuple as parameter method

c8fb1e2

record.py: simulation parameters to _plot_animate passed with a named…

597409e

…tuple

minor errors (due to tragic namedtuple

cf742d9

Merge remote-tracking branch 'origin/models_igr'

ef7ae7f

# Conflicts: # neuronpp/utils/record.py

fixed Record to work with markers, not work yet

79f38d8

Corrected errors in record.py; changed its parameters a bit

34a7336

Merge remote-tracking branch 'origin/models_igr' into models_igr

387e21a

# Conflicts: # neuronpp/utils/record.py

minor errors in record.py, plot_animate()

f2d346e

minor errors in inheritance in record.py

8411e81

Merge remote-tracking branch 'origin/master' into models_igr

2ca9dec

changes to Record

be385e8

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

branch: models_igr: graph display + record selection #6

branch: models_igr: graph display + record selection #6

igorpodolak commented Mar 23, 2020 •

edited

Loading

ziemowit-s Mar 23, 2020

igorpodolak Mar 23, 2020

ziemowit-s Mar 23, 2020

igorpodolak Mar 23, 2020

ziemowit-s Mar 24, 2020

ziemowit-s Mar 23, 2020

igorpodolak Mar 23, 2020

ziemowit-s Mar 24, 2020

ziemowit-s Mar 23, 2020

igorpodolak Mar 23, 2020

wsredniawa Mar 23, 2020

igorpodolak Mar 23, 2020

igorpodolak Mar 23, 2020

ziemowit-s Mar 24, 2020

ziemowit-s Mar 24, 2020

ziemowit-s Mar 26, 2020

ziemowit-s Mar 26, 2020

igorpodolak Mar 27, 2020

igorpodolak Mar 27, 2020

ziemowit-s Mar 26, 2020

igorpodolak Mar 27, 2020

ziemowit-s Mar 26, 2020

igorpodolak Mar 27, 2020

ziemowit-s Mar 26, 2020

	def _class_tcks(self, label, true_class, pred_class, t, r, stepsize, dt):
	def _draw_true_predicted_class_marks(self, label, true_class, pred_class, t, r, stepsize, dt):

	y_pos -= 6
	y_pos -= self.population_sizes['hid_1'] // 2 + self.population_sizes[pop_name]

		true_y = [min(r) + 2] * len(true_x)
		pred_y = [max(r) - 2] * len(pred_x)

branch: models_igr: graph display + record selection #6

Are you sure you want to change the base?

branch: models_igr: graph display + record selection #6

Conversation

igorpodolak commented Mar 23, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

igorpodolak commented Mar 23, 2020 •

edited

Loading