Skip to content

Commit

Permalink
tidy up
Browse files Browse the repository at this point in the history
  • Loading branch information
@oliviaguest
oliviaguest committed May 4, 2019
1 parent 3ae3954 commit cbdc31c
Show file tree
Hide file tree
Showing 3 changed files with 105 additions and 102 deletions.
Empty file added shapes_experiment/accuracy/.gitkeep
View file
Empty file.
49 changes: 26 additions & 23 deletions shapes_experiment/exemplar_model_circle_square.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -155,8 +155,12 @@ def create_and_save_exemplar_model(
# Load the representations for all inputs for a single layer:
layer_filename = EXP_DIR + 'layer_representations/' + \
stim_dir + layer_name.replace("/", "_") + '.csv'
df = pd.read_csv(layer_filename)

try:
df = pd.read_csv(layer_filename)
except FileNotFoundError:
print('You need to run the deep network before this to get the'
' layer representations for the stimuli!')
exit()
# Create all category combinations:
prototypes_a, categories_a, prototypes_b, categories_b = \
create_all_categories(list(df.columns))
Expand Down Expand Up @@ -252,9 +256,9 @@ def create_and_save_exemplar_model(
max_accuracy.append(
np.sum(exemplar_model['Max Correct']) /
exemplar_model['Max Correct'].count())
print(prototype_a)
# print(prototype_a)
exemplar_model = get_optimum_correct(exemplar_model, items_a, items_b)
print(exemplar_model)
# print(exemplar_model)

# print('Luce choice accuracy at', layer_name,
# ':', accuracy[layer_index],)
Expand Down Expand Up @@ -320,25 +324,24 @@ def create_and_save_exemplar_model(
' Correct']]
.mean())

accuracy_df.set_value(layer_name, 'Luce Accuracy', np.mean(
accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
accuracy_df.set_value(layer_name, 'Max Accuracy', np.mean(max_accuracy[layer_index * (
len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
accuracy_df.set_value(layer_name, 'Optimum Accuracy', np.mean(
optimum_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
accuracy_df.set_value(layer_name, 'Same Hue Luce Accuracy', np.mean(
same_hue_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
accuracy_df.set_value(layer_name, 'Same Size Luce Accuracy', np.mean(
same_size_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
accuracy_df.set_value(layer_name, 'Same Shape Luce Accuracy', np.mean(
same_shape_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
accuracy_df.set_value(layer_name, 'Same Hue Optimum Accuracy', np.mean(
optimum_same_hue_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
accuracy_df.set_value(layer_name, 'Same Size Optimum Accuracy', np.mean(
optimum_same_shape_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
accuracy_df.set_value(layer_name, 'Same Shape Optimum Accuracy', np.mean(
optimum_same_size_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))]))
print(accuracy_df)
accuracy_df.at[layer_name, 'Luce Accuracy'] = np.mean(
accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])
accuracy_df.at[layer_name, 'Max Accuracy'] = np.mean(max_accuracy[layer_index * (
len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])
accuracy_df.at[layer_name, 'Optimum Accuracy'] = np.mean(
optimum_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])
accuracy_df.at[layer_name, 'Same Hue Luce Accuracy'] = np.mean(
same_hue_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])
accuracy_df.at[layer_name, 'Same Size Luce Accuracy'] = np.mean(
same_size_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])
accuracy_df.at[layer_name, 'Same Shape Luce Accuracy'] = np.mean(
same_shape_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])
accuracy_df.at[layer_name, 'Same Hue Optimum Accuracy'] = np.mean(
optimum_same_hue_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])
accuracy_df.at[layer_name, 'Same Size Optimum Accuracy'] = np.mean(
optimum_same_shape_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])
accuracy_df.at[layer_name, 'Same Shape Optimum Accuracy'] = np.mean(
optimum_same_size_accuracy[layer_index * (len(categories_a[0])):(layer_index + 1) * (len(categories_a[0]))])


accuracy = np.asarray(accuracy) \
Expand Down
158 changes: 79 additions & 79 deletions shapes_experiment/exemplar_model_circle_square_bounding_box.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -171,7 +171,7 @@ def create_all_categories(items):
index=LAYER_NAMES)
accuracy_df['Stimulus Type'] = stim_type
accuracy_df['Hue Type'] = hue
print(hue, stim_type)
# print(hue, stim_type)

results_dir = EXP_DIR + '/results/' + stim_dir
figures_dir = EXP_DIR + '/figures/' + stim_dir
Expand All @@ -193,11 +193,11 @@ def create_all_categories(items):
'_' + layer_name.replace("/", "_")

# Load the representations for all inputs for a single layer:
layer_filename = (EXP_DIR + 'layer_representations/'
+ stim_dir + layer_name.replace("/", "_")
+ '.csv')
layer_filename = (EXP_DIR + 'layer_representations/' +
stim_dir + layer_name.replace("/", "_") +
'.csv')
print('\tOpening CSV file with layer representations...')
print(layer_filename)
# print(layer_filename)
try:
df = pd.read_csv(layer_filename)
except FileNotFoundError:
Expand Down Expand Up @@ -248,18 +248,18 @@ def create_all_categories(items):
similarity_to_B_prototype = A_items_to_B_prototype.append(
B_items_to_B_prototype)
similarity_to_both = \
pd.DataFrame(similarity_to_A_prototype[prototype_a] +
similarity_to_B_prototype[prototype_b])
pd.DataFrame(similarity_to_A_prototype[prototype_a]
+ similarity_to_B_prototype[prototype_b])

# Probability that an item is in A
prob_A = pd.DataFrame(
similarity_to_A_prototype[prototype_a]
/ similarity_to_both[0], columns=['Probability A'])
similarity_to_A_prototype[prototype_a] /
similarity_to_both[0], columns=['Probability A'])

# Probability that an item is in B
prob_B = pd.DataFrame(
similarity_to_B_prototype[prototype_b]
/ similarity_to_both[0], columns=['Probability B'])
similarity_to_B_prototype[prototype_b] /
similarity_to_both[0], columns=['Probability B'])

exemplar_model = pd.concat(
[prob_A, prob_B], axis=1, join='inner')
Expand Down Expand Up @@ -305,13 +305,13 @@ def create_all_categories(items):
exemplar_model.loc[items_b, 'Max Correct'] = \
exemplar_model['Probability B'].loc[items_b] > \
exemplar_model['Probability A'].loc[items_b]
max_accuracy.append(np.sum(exemplar_model['Max Correct'])
/ exemplar_model['Max Correct'].count())
max_accuracy.append(np.sum(exemplar_model['Max Correct']) /
exemplar_model['Max Correct'].count())

print(prototype_a)
# print(prototype_a)
exemplar_model = get_optimum_correct(
exemplar_model, items_a, items_b)
print(exemplar_model)
# print(exemplar_model)

create_and_save_figures(exemplar_model, figures_base_filename,
layer_index, layer_name, show=False)
Expand All @@ -322,102 +322,102 @@ def create_all_categories(items):
exemplar_model = pd.concat([exemplar_model, dim_df], axis=1,
join='inner')

hue_accuracy.append(exemplar_model[exemplar_model['Hue'] !=
exemplar_model[
hue_accuracy.append(exemplar_model[exemplar_model['Hue']
!= exemplar_model[
'Prototype Hue']]
[['Probability Correct']].mean())
print(exemplar_model['Hue'])
print(exemplar_model['Prototype Hue'])
# print(exemplar_model['Hue'])
# print(exemplar_model['Prototype Hue'])

shape_accuracy.append(exemplar_model[exemplar_model['Shape'] !=
exemplar_model
shape_accuracy.append(exemplar_model[exemplar_model['Shape']
!= exemplar_model
['Prototype Shape']][[
'Probability'
' Correct']].mean())

size_accuracy.append(exemplar_model[exemplar_model['Size'] !=
exemplar_model
size_accuracy.append(exemplar_model[exemplar_model['Size']
!= exemplar_model
['Prototype Size']][[
'Probability'
' Correct']].mean())

same_hue_accuracy.append(exemplar_model[exemplar_model
['Hue'] ==
exemplar_model
['Hue']
== exemplar_model
['Prototype Hue']]
[['Probability'
' Correct']].mean())

same_shape_accuracy.append(exemplar_model[exemplar_model
['Shape'] ==
exemplar_model
['Shape']
== exemplar_model
['Prototype'
' Shape']]
[['Probability Correct']].mean())

same_size_accuracy.append(exemplar_model[exemplar_model
['Size'] ==
exemplar_model
['Size']
== exemplar_model
['Prototype'
' Size']][[
'Probability'
' Correct']]
.mean())

optimum_accuracy.append(np.sum(
exemplar_model['Optimum Correct']) /
exemplar_model['Optimum Correct'].count())
exemplar_model['Optimum Correct'])
/ exemplar_model['Optimum Correct'].count())

optimum_same_hue_accuracy.append(
exemplar_model[exemplar_model
['Hue'] ==
exemplar_model
['Hue']
== exemplar_model
['Prototype Hue']]
[['Optimum'
' Correct']].mean())

optimum_same_shape_accuracy.append(
exemplar_model[exemplar_model
['Shape'] ==
exemplar_model
['Shape']
== exemplar_model
['Prototype'
' Shape']]
[['Optimum Correct']].mean())

optimum_same_size_accuracy.append(
exemplar_model[exemplar_model
['Size'] ==
exemplar_model
['Size']
== exemplar_model
['Prototype'
' Size']][[
'Optimum'
' Correct']]
.mean())
layer_slice = slice(layer_index * (len(items_a) + len(items_b)),
(layer_index + 1) * (len(items_a) +
len(items_b)))
accuracy_df.set_value(layer_name, 'Luce Accuracy', np.mean(
accuracy[layer_slice]))
accuracy_df.set_value(layer_name, 'Max Accuracy', np.mean(
max_accuracy[layer_slice]))
accuracy_df.set_value(layer_name, 'Optimum Accuracy', np.mean(
optimum_accuracy[layer_slice]))
accuracy_df.set_value(layer_name, 'Same Hue Luce Accuracy',
np.mean(same_hue_accuracy[layer_slice]))
accuracy_df.set_value(layer_name, 'Same Size Luce Accuracy',
np.mean(same_size_accuracy[layer_slice]))
accuracy_df.set_value(layer_name, 'Same Shape Luce Accuracy',
np.mean(same_shape_accuracy[layer_slice]))
accuracy_df.set_value(layer_name, 'Same Hue Optimum Accuracy',
np.mean(
optimum_same_hue_accuracy[layer_slice]))
accuracy_df.set_value(layer_name, 'Same Size Optimum Accuracy',
np.mean(
optimum_same_shape_accuracy[
layer_slice]))
accuracy_df.set_value(layer_name, 'Same Shape Optimum Accuracy',
np.mean(
optimum_same_size_accuracy[layer_slice]))
(layer_index + 1) * (len(items_a)
+ len(items_b)))
accuracy_df.at[layer_name, 'Luce Accuracy'] = np.mean(
accuracy[layer_slice])
accuracy_df.at[layer_name, 'Max Accuracy'] = np.mean(
max_accuracy[layer_slice])
accuracy_df.at[layer_name, 'Optimum Accuracy'] = np.mean(
optimum_accuracy[layer_slice])
accuracy_df.at[layer_name, 'Same Hue Luce Accuracy'] =\
np.mean(same_hue_accuracy[layer_slice])
accuracy_df.at[layer_name, 'Same Size Luce Accuracy'] =\
np.mean(same_size_accuracy[layer_slice])
accuracy_df.at[layer_name, 'Same Shape Luce Accuracy'] =\
np.mean(same_shape_accuracy[layer_slice])
accuracy_df.at[layer_name, 'Same Hue Optimum Accuracy'] =\
np.mean(
optimum_same_hue_accuracy[layer_slice])
accuracy_df.at[layer_name, 'Same Size Optimum Accuracy'] =\
np.mean(
optimum_same_shape_accuracy[
layer_slice])
accuracy_df.at[layer_name, 'Same Shape Optimum Accuracy'] =\
np.mean(
optimum_same_size_accuracy[layer_slice])
layer_shape = (len(LAYER_NAMES), len(items_a) + len(items_b))
accuracy = np.asarray(accuracy).reshape(layer_shape).mean(axis=1)
max_accuracy = np.asarray(max_accuracy).\
Expand All @@ -435,21 +435,21 @@ def create_all_categories(items):
same_shape_accuracy = np.asarray(same_shape_accuracy).\
reshape(layer_shape).mean(axis=1)

accuracy_df.to_csv(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'accuracy.csv')
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'luce.csv', accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'max.csv', max_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'hue_luce.csv', hue_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'shape_luce.csv', shape_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'size_luce.csv', size_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'same_hue_luce.csv', same_hue_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'same_shape_luce.csv', same_shape_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_")
+ 'same_size_luce.csv', same_size_accuracy)
accuracy_df.to_csv(ACCURACY_DIR + stim_dir.replace("/", "_") +
'accuracy.csv')
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_") +
'luce.csv', accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_") +
'max.csv', max_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_") +
'hue_luce.csv', hue_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_") +
'shape_luce.csv', shape_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_") +
'size_luce.csv', size_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_") +
'same_hue_luce.csv', same_hue_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_") +
'same_shape_luce.csv', same_shape_accuracy)
np.savetxt(ACCURACY_DIR + stim_dir.replace("/", "_") +
'same_size_luce.csv', same_size_accuracy)

0 comments on commit cbdc31c

Please sign in to comment.