def embed_motifs(self, bg_seqs, pwms):
print("embedding motifs...")
embedded_seqs = []
pwm = pwms[0]
locs = []
for seq in bg_seqs:
mf_instance = mf.sample_motif(pwm, random_rev_comp=False)
loc = np.random.randint(0, 500)
locs.append(loc)
embedded_seqs.append(mf.replace_at(seq, mf_instance, loc))
utils.plot_hist(locs)
return embedded_seqs
def _calc_ranges(self, _ev=None):
values = self.map_image.get_bands(self.map_window.channels_histogram)
self.x_range, self.y_range = ([values[i].min(), values[i].max()]
for i in range(2))
for i in range(2):
values[i][values[i] < values[i].min() + 0.00000001] = np.nan
self.graphs = [plot_hist(values[i]) for i in range(2)]
cnt[ img_arr[i, j] ] += 1
return cnt[:-1], intensity
def intensity_hist(img_arr, n_bins=256):
bins = np.arange(n_bins + 1)
cnt, intensity = poy_histogram(img_arr, bins=bins)
return cnt, intensity[:-1]
L = 256
sample2_cnt, sample2_intensity = intensity_hist(sample2_arr, n_bins=L)
result3_cnt, result3_intensity = intensity_hist(result3_arr, n_bins=L)
result4_cnt, result4_intensity = intensity_hist(result4_arr, n_bins=L)
utils.plot_hist("prob2c",
[sample2_intensity, sample2_cnt],
[result3_intensity, result3_cnt],
[result4_intensity, result4_cnt]
)
# prob (d)
def poy_cumsum(arr):
res = []
cumsum = 0
for a in arr:
cumsum += a
res.append(cumsum)
res = np.array(res)
return res
def global_hist_equal(img_arr):
'''
epoch_train_loss.append(np.mean(train_losses))
epoch_val_loss.append(np.mean(val_losses))
epoch_train_dsc.append(np.mean(train_dsc))
epoch_val_dsc.append(np.mean(val_dsc))
early_stopping(np.average(val_losses), model)
if early_stopping.early_stop:
print("Early stopping at epoch: ", epoch)
break
print('='*30)
print('Average DSC score =', np.array(val_dsc).mean())
utils.plot_hist(epoch, np.array(epoch_train_loss), np.array(epoch_val_loss), "Loss")
utils.plot_hist(epoch, np.array(epoch_train_dsc), np.array(epoch_val_dsc), "DSC Score")
# check model outputs on validation data
model.eval()
idx = np.random.randint(0,batch_size)
val_dsc = []
with torch.no_grad():
for x_val, y_val in val_loader:
x_val, y_val = x_val.to(device), y_val.to(device)
preds = model(x_val)
dsc = losses.dice_score(preds, y_val)
val_dsc.append(dsc/x_val.shape[0])
batch_size=args.batch_size,
shuffle=False,
pin_memory=True,
num_workers=8)
# random search
start = time.time()
best_acc = 0.0
acc_list = list()
best_choice = list()
for epoch in range(args.random_search):
choice = utils.random_choice(args.num_choices, args.layers)
top1_acc = validate(args,
epoch,
val_loader,
device,
model,
criterion,
super=True,
choice=choice)
acc_list.append(top1_acc)
if best_acc < top1_acc:
best_acc = top1_acc
best_choice = choice
print('acc_list:')
for i in acc_list:
print(i)
print('best_acc:{} \nbest_choice:{}'.format(best_acc, best_choice))
utils.plot_hist(acc_list, name=args.exp_name)
utils.time_record(start)
'[%d]/[%d] Train Loss:%.4f\t Train Acc:%.4f\t Val Loss:%.4f\t Val Acc: %.4f'
% (epoch + 1, epochs, np.mean(train_loss), np.mean(train_acc),
np.mean(val_loss), np.mean(val_acc)))
epoch_train_loss.append(np.mean(train_loss))
epoch_val_loss.append(np.mean(val_loss))
epoch_train_acc.append(np.mean(train_acc))
epoch_val_acc.append(np.mean(val_acc))
early_stopping(np.average(val_loss), model)
if early_stopping.early_stop:
print("Early stopping at epoch: ", epoch)
break
utils.plot_hist(epoch, np.array(epoch_train_loss), np.array(epoch_val_loss),
"Loss")
plt.title('Lowest loss = %.4f' % epoch_val_loss[-1])
utils.plot_hist(epoch, np.array(epoch_train_acc), np.array(epoch_val_acc),
"Acc")
plt.title('Best accuracy = %.4f' % epoch_val_acc[-1])
# Load up the best model and view preds
model.load_state_dict(torch.load('checkpoint.pt'))
outputs = []
def hook(module, input, output):
outputs.append(output)
from keras.layers import Dense
import sys
sys.path.append("../")
from utils import load_mnist_1D
from utils import save_model_viz, save_weights, save_hist, plot_hist
RUN_ID = 'mlp'
(x_train, y_train), (x_test, y_test) = load_mnist_1D()
model = Sequential()
model.add(Dense(64, input_shape=(784, ), activation='relu'))
model.add(Dense(10, activation='softmax'))
model.compile(optimizer='adam',
loss='categorical_crossentropy',
metrics=['accuracy'])
save_model_viz(RUN_ID, model)
hist = model.fit(x_train,
y_train,
epochs=5,
batch_size=32,
verbose=1,
validation_data=(x_test, y_test))
save_weights(RUN_ID, model)
save_hist(RUN_ID, hist)
plot_hist(RUN_ID)
y_aug = np.array(y_new)
# ====== print info ====== #
print("Train set:", x_train.shape, y_train.shape)
print("Valid set:", x_valid.shape, y_valid.shape)
print("Test set:", x_test.shape, y_test.shape)
print("Augmented training set:", x_aug.shape, y_aug.shape)
# ====== checking distribution of train, valid, test matching ====== #
train_dist = itemfreq(y_train)
valid_dist = itemfreq(y_valid)
test_dist = itemfreq(y_test)
plt.figure()
ax = plt.subplot(3, 1, 1)
plot_hist(y_train, ax, "Training distribution")
ax = plt.subplot(3, 1, 2)
plot_hist(y_valid, ax, "Validating distribution")
ax = plt.subplot(3, 1, 3)
plot_hist(y_test, ax, "Testing distribution")
plt.tight_layout()
# ====== convert labels to one_hot for training ====== #
labels = ["Number: %d" % i for i in y_train[:16]]
y_train = one_hot(y_train, nb_classes=10)
y_aug = one_hot(y_aug, nb_classes=10)
y_test = one_hot(y_test, nb_classes=10)
y_valid = one_hot(y_valid, nb_classes=10)
plt.figure()
plt.imshow(y_train[:16], cmap=plt.cm.Greys_r)
plt.xticks(np.arange(10))
def compute_deeplift_scores(
TARGET_DIRECTORY, dataset, X, Y,
keras_model_file, reference_label, non_reference_label, base_neuron_label,
mask, gpu_id, dropout, threshold, percentage_cutoff, cluster_mask,
flags):
"""
Wrapper function for model reduction, called by main.py
Uses DeepLIFT to compute saliency scores for feature selection, with the average data used as reference
See https://github.com/kundajelab/deeplift for DeepLIFT implementation
Inputs:
- TARGET_DIRECTORY: general directory path to write files to (str)
- dataset: choice of dataset, along with seed and fold number (str)
- X: Numpy array containing data matrices
- Y: Numpy array containing data labels
- keras_model_file: name of existing model file (str)
- reference_label: how the reference class is represented in Y, usually 0 (int)
- non_reference_label: how the other class(es) is (are) represented in Y, usually 1 (int)
- base_neuron_label: label to be used as the base, usually 0 (int)
- mask: Numpy array, usually initialised as all 1s unless neurons are repeatedly removed
- gpu_id: ID of GPU to use (int)
- dropout: fraction of neurons to turn off (float)
- threshold: usually set as 1.0, represents the previous percentage_cutoff when repeatedly removing neurons (float)
- percentage_cutoff: usually set as 0.95 to keep 5% of the most significant features (float)
- cluster_mask: Numpy array containing mask obtained from CLIP
- flags: used to vary model settings, see main.py (dict)
Returns:
- new_model_file: directory path to the new model (str)
- mask_2D_flattened: Numpy array of 1s and 0s, with 1 representing a selected feature
"""
keras_model = keras.models.load_model(keras_model_file)
print(keras_model.summary()) # original model
deeplift_model = kc.convert_model_from_saved_files(
keras_model_file,
nonlinear_mxts_mode=deeplift.layers.NonlinearMxtsMode.DeepLIFT_GenomicsDefault)
print(deeplift_model.get_layers())
mode = 'average'
X_masked, mapping = get_masked_data(X, mask)
Y = np.argmax(Y, axis=1)
reference = get_reference(mode, reference_label, X_masked, Y)
print('+++++++++++++ Computing DeepLIFT scores ++++++++++++++')
print('previous threshold', threshold, 'new threshold', percentage_cutoff)
find_scores_layer_idx = 0
input_scores = np.zeros(X_masked[Y == non_reference_label].shape)
layer_scores = []
task_id = base_neuron_label
for layer_idx, layer in enumerate(deeplift_model.get_layers()):
if type(layer).__name__ == 'Dense' or type(layer).__name__ == 'Input':
deeplift_contribs_func = deeplift_model.get_target_contribs_func(find_scores_layer_idx=layer_idx, target_layer_idx=-2)
scores = np.array(deeplift_contribs_func(task_idx=task_id,
input_references_list=reference,
input_data_list=[X_masked[Y == non_reference_label]],
batch_size=10,
progress_update=50))
sum_scores = np.zeros(scores.shape[1])
for score in scores:
sum_scores += score
sum_scores = np.absolute(sum_scores)
if sum_scores.shape[0] > 2:
plot_hist(TARGET_DIRECTORY, sum_scores, dataset + '_t_' + str(threshold) + '_layer_' + str(layer_idx))
print('layer', layer_idx, 'type is: ', type(layer), 'scores dimensions are: ', scores.shape, 'sum_scores', sum_scores.shape)
layer_scores.append(sum_scores)
if layer_idx == 0:
input_scores = np.square(scores)
layer_scores.append([])
elif type(layer).__name__ == 'NoOp' or type(layer).__name__ == 'Softmax':
layer_scores.append([])
print('layer', layer_idx, 'type is: ', type(layer).__name__)
alpha = (1 - (percentage_cutoff/threshold))
new_model, mask_2D = compute_new_reduced_model(keras_model, dropout, layer_scores, 2, alpha, mapping, cluster_mask, flags)
input_sum_scores = np.zeros(X_masked.shape[1])
for input_score in input_scores:
input_sum_scores += input_score
padded_sum_scores = get_padded_data(input_sum_scores, mapping)
full_matrix = create_matrix(padded_sum_scores)
mkdir(TARGET_DIRECTORY + './important_features/')
np.savetxt(
TARGET_DIRECTORY + './important_features/' + dataset + '_scores_deeplift_reduced_r_' + str(threshold) + '_t_' + str(percentage_cutoff) + '.csv',
np.transpose(np.array(input_scores)), delimiter= ',')
np.savetxt(
TARGET_DIRECTORY + './important_features/' + dataset + '_scores_reshaped_reduced_r_' + str(threshold) + '_t_' + str(percentage_cutoff) + '.csv',
full_matrix, delimiter=",")
np.savetxt(
TARGET_DIRECTORY + './important_features/' + dataset + '_deeplift_features_nodes_r_' + str(threshold) + '_t_' + str(percentage_cutoff) + '.csv',
mask_2D)
mkdir(TARGET_DIRECTORY + './reduced_models/')
new_model_file = TARGET_DIRECTORY + './reduced_models/' + dataset + '_from_' + str(threshold) + '_to_' + str(percentage_cutoff) + '.h5'
new_model.save(new_model_file)
os.remove(keras_model_file)
mask_2D_flattened = corr_mx_flatten_single(mask_2D)
return new_model_file, mask_2D_flattened
def save_and_plot_stats_environment(self, run_num):
"""Saves all environment-relevant statistics for later analysis."""
run_subfolder = "game_" + str(run_num) + "/"
os.makedirs(self.plot_directory + run_subfolder + "turtle_dynamics/")
self.exp_details()
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'agent_act_list.csv',
self.agent_act_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'human_act_list.csv',
self.human_act_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'action_timesteps.csv',
self.action_timesteps,
delimiter=',',
fmt='%f')
human_act_list = np.genfromtxt(self.plot_directory + run_subfolder +
"turtle_dynamics/" +
'human_act_list.csv',
delimiter=',')
agent_act_list = np.genfromtxt(self.plot_directory + run_subfolder +
"turtle_dynamics/" +
'agent_act_list.csv',
delimiter=',')
plot_hist(
agent_act_list, self.plot_directory + run_subfolder +
"turtle_dynamics/" + 'agent_act_hist_'
"game_" + str(run_num), 'Agent Action Histogram')
plot_hist(
human_act_list, self.plot_directory + run_subfolder +
"turtle_dynamics/" + 'human_act_hist_'
"game_" + str(run_num), 'Human Action Histogram')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'turtle_pos_x.csv',
self.turtle_pos_x,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'turtle_pos_y.csv',
self.turtle_pos_y,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'time_turtle_pos.csv',
self.time_turtle_pos,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'turtle_vel_x.csv',
self.turtle_vel_x,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'turtle_vel_y.csv',
self.turtle_vel_y,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'time_turtle_vel.csv',
self.time_turtle_vel,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'turtle_accel_x.csv',
self.turtle_acc_x,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'turtle_accel_y.csv',
self.turtle_acc_y,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'time_turtle_acc.csv',
self.time_turtle_acc,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'real_act_list.csv',
self.real_act_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'time_real_act_list.csv',
self.time_real_act_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'used_human_act_list.csv',
self.used_human_act_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + run_subfolder + "turtle_dynamics/" +
'used_human_act_time_list.csv',
self.used_human_act_time_list,
delimiter=',',
fmt='%f')
subplot(self.plot_directory + run_subfolder + "turtle_dynamics/",
self.turtle_pos_x, self.turtle_vel_x, self.turtle_acc_x,
self.time_turtle_pos, self.time_turtle_vel,
self.time_turtle_acc, human_act_list, self.action_timesteps,
"x", control_mode)
subplot(self.plot_directory + run_subfolder + "turtle_dynamics/",
self.turtle_pos_y, self.turtle_vel_y, self.turtle_acc_y,
self.time_turtle_pos, self.time_turtle_vel,
self.time_turtle_acc, agent_act_list, self.action_timesteps,
"y", control_mode)
plt.figure("Real_Human_Actions_Comparison", figsize=(25, 10))
plt.grid()
plt.ylabel('Human Actions')
plt.xlabel('Msg Timestamp(seconds)')
# plt.xticks(plt.xticks(np.arange(min(min(time_real_act_list),min(self.used_human_act_time_list)), max(max(time_real_act_list),max(used_human_act_time_list)), 1)))
plt.scatter(self.time_real_act_list, self.real_act_list)
plt.scatter(self.used_human_act_time_list, self.used_human_act_list)
plt.savefig(self.plot_directory + run_subfolder +
"human_real_action_comparison",
dpi=150)
self.reset_lists()
def save_and_plot_stats_rl(self):
"""Saves & plots all the rl-relevant statistics for later analysis."""
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'alpha_values.csv',
self.alpha_values,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'policy_loss.csv',
self.policy_loss_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'value_loss.csv',
self.value_loss_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'q_loss.csv',
self.q_loss_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'rewards_list.csv',
self.rewards_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'turn_list.csv',
self.turn_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'means.csv',
self.mean_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'stdev.csv',
self.stdev_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' +
'critics_lr_list.csv',
self.critics_lr_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' +
'value_critic_lr_list.csv',
self.value_critic_lr_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'actor_lr_list.csv',
self.actor_lr_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'rl_dynamics/' + 'trials_list.csv',
self.trials_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'agent_act_list_total.csv',
self.agent_act_list_total,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'human_act_list_total.csv',
self.human_act_list_total,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'human_action_delay_list.csv',
self.human_action_delay_list,
delimiter=',',
fmt='%f')
np.savetxt(self.plot_directory + 'fps_list.csv',
self.fps_list,
delimiter=',',
fmt='%f')
plot_hist(self.fps_list, self.plot_directory,
'fps_list_' + str(self.game.fps))
np.savetxt(self.plot_directory + 'exec_time_list.csv',
self.exec_time_list,
delimiter=',',
fmt='%f')
plot(range(len(self.alpha_values)),
self.alpha_values,
"alpha_values",
'Alpha Value',
'Number of Gradient Updates',
self.plot_directory,
save=True)
plot(range(len(self.policy_loss_list)),
self.policy_loss_list,
"policy_loss",
'Policy loss',
'Number of Gradient Updates',
self.plot_directory,
save=True)
plot(range(len(self.value_loss_list)),
self.value_loss_list,
"value_loss_list",
'Value loss',
'Number of Gradient Updates',
self.plot_directory,
save=True)
plot(range(len(self.rewards_list)),
self.rewards_list,
"Rewards_per_game",
'Total Rewards per Game',
'Number of Games',
self.plot_directory,
save=True)
plot(range(len(self.turn_list)),
self.turn_list,
"Steps_per_game",
'Turns per Game',
'Number of Games',
self.plot_directory,
save=True)
plot(range(len(self.critics_lr_list)),
self.critics_lr_list,
"critics_lr_list",
'Critic lr',
'Number of Gradient Updates',
self.plot_directory,
save=True)
plot(range(len(self.value_critic_lr_list)),
self.value_critic_lr_list,
"value_critic_lr_list",
'Value lr',
'Number of Gradient Updates',
self.plot_directory,
save=True)
plot(range(len(self.actor_lr_list)),
self.actor_lr_list,
"actor_lr_list",
'Actor lr',
'Number of Gradient Updates',
self.plot_directory,
save=True)
try:
plot(range(UPDATE_INTERVAL, MAX_STEPS + UPDATE_INTERVAL,
UPDATE_INTERVAL),
self.mean_list,
"trials",
'Tests Score',
'Number of Interactions',
self.plot_directory,
save=True,
variance=True,
stdev=self.stdev_list)
except:
print("Trials did not ploted")
plot_hist(self.agent_act_list_total,
self.plot_directory + 'agent_act_hist_total',
'Agent Action Histogram')
plot_hist(self.human_act_list_total,
self.plot_directory + 'human_act_hist_total',
'Human Action Histogram')
# human_action_delay_list_new = [elem for elem in self.human_action_delay_list if elem <0.8] # remove outliers caused by saving and plotting functions
plot_hist(self.human_action_delay_list,
self.plot_directory + 'human_action_delay_list',
'Human Action Delay Histogram')