def make_plot(data):
fig, axs = plt.subplots(2, 3, figsize=(8, 6), sharey=True)
for i, taxon in enumerate(data):
if i <= 2:
ax = axs[0][i]
else:
ax = axs[1][i - 3]
xs = data[taxon][0]
ys = data[taxon][1]
positions = data[taxon][2]
colors = []
for position in positions:
if position == "T": colors.append("red")
if position == "M": colors.append("magenta")
if position == "B": colors.append("cyan")
ax.scatter(xs, ys, alpha=0.6, c=colors, edgecolor='k')
# line of best fit
grid = np.r_[0.5:5:512j]
k0 = smooth.NonParamRegression(xs,
ys,
method=npr_methods.SpatialAverage())
k0.fit()
ax.plot(grid, k0(grid), "k", linewidth=2)
test = stats.pearsonr(data[taxon][0], data[taxon][1])
print taxon, test
ax.set_title(taxon.split(";")[-1], fontsize=12)
ax.grid(ls="--")
ax.set_xticks([0, 1, 2, 3, 4, 5])
# if i!=0:
# ax.set_yticklabels([])
fig.add_subplot(111, frameon=False)
plt.tick_params(labelcolor='none',
top=False,
bottom=False,
left=False,
right=False)
plt.xlabel("Distance to surface (cm)")
plt.ylabel("Standardized abundance of taxon")
plt.tight_layout()
plt.savefig("figure.png", dpi=300)
def aseErr_expression(genes_fpkm,called_ase_file,true_ase_file):
called_ase_dict = loadASE_list(called_ase_file)
true_ase_dict = loadASE_list(true_ase_file)
gene_expression = loadFPKM(genes_fpkm)
##to be filled in with error values later (not paired with names)
ase_errors = []
##filled with fpkms of expressed genes, in step with ase_errors
expression_values = []
##called_ase genes are a subset of true_ase genes
for ensid in called_ase_dict.keys():
true_ase = true_ase_dict[ensid]
called_ase = called_ase_dict[ensid]
fpkm = gene_expression[ensid]
true_ase = math.log( (true_ase/(1-true_ase)),2)
called_ase = math.log( (called_ase/(1-called_ase)),2)
expression_values.append(math.log(fpkm,2))
ase_err = math.fabs(true_ase - called_ase)
ase_errors.append(ase_err)
plt.scatter(expression_values,ase_errors,alpha=.3,marker='o')
r,pval = stats.spearmanr(expression_values,ase_errors)
plt.xlabel('log2(FPKM)\nr=%f'%r)
plt.ylabel('log2 ASE_errors')
plt.ylim([0,2.5])
plt.xlim([-1,12])
k0 = smooth.NonParamRegression(expression_values, ase_errors, method=npr_methods.SpatialAverage())
k0.fit()
xs = np.arange(-1,12,.01)
plt.plot(xs, k0(xs), linewidth=2)
plt.show()
def results(dataset, metadata_path, w2v, rescale=None):
print("Configuring Tensorflow Graph")
with tf.Graph().as_default():
sess, siamese_model = initialize_tf_graph(metadata_path, w2v)
dataset.test.open()
dataset.train.open()
avg_test_loss, avg_test_pco, test_result_set = evaluate(
sess=sess,
dataset=dataset.test,
model=siamese_model,
step=-1,
max_dev_itr=0,
mode='test')
avg_train_loss, avg_train_pco, train_result_set = evaluate(
sess=sess,
dataset=dataset.train,
model=siamese_model,
max_dev_itr=0,
step=-1,
mode='train')
dataset.test.close()
dataset.train.close()
print('TEST RESULTS:\nMSE: {}\t Pearson Correlation: {}\n\n'
'TRAIN RESULTS:\nMSE: {}\t Pearson Correlation: {}'.format(
avg_test_loss, avg_test_pco, avg_train_loss, avg_train_pco))
_, _, train_sims, train_gt = train_result_set
_, _, test_sims, test_gt = test_result_set
grid = np.r_[0:1:1000j]
if rescale is not None:
train_gt = datasets.rescale(train_gt,
new_range=rescale,
original_range=[0.0, 1.0])
test_gt = datasets.rescale(test_gt,
new_range=rescale,
original_range=[0.0, 1.0])
# grid = np.r_[rescale[0]:rescale[1]:1000j]
figure_path = os.path.join(siamese_model.exp_dir,
'results_test_sim.jpg')
reg_fig_path = os.path.join(siamese_model.exp_dir,
'results_line_fit.jpg')
plt.title('Regression Plot for Test Set Similarities')
plt.ylabel('Ground Truth Similarities')
plt.xlabel('Predicted Similarities')
print("Performing Non Parametric Regression")
non_param_reg = non_parametric_regression(
train_sims, train_gt, method=npr_methods.SpatialAverage())
reg_test_sim = non_param_reg(test_sims)
reg_pco = pearsonr(reg_test_sim, test_gt)
reg_mse = mean_squared_error(test_gt, reg_test_sim)
print("Post Regression Test Results:\nPCO: {}\nMSE: {}".format(
reg_pco, reg_mse))
plt.scatter(reg_test_sim, test_gt, label='Similarities', s=0.2)
plt.savefig(figure_path)
plt.clf()
plt.title('Regression Plot for Test Set Similarities')
plt.ylabel('Ground Truth Similarities')
plt.xlabel('Predicted Similarities')
plt.scatter(test_sims, test_gt, label='Similarities', s=0.2)
plt.plot(grid,
non_param_reg(grid),
label="Local Linear Smoothing",
linewidth=2.0,
color='r')
plt.savefig(reg_fig_path)
print("saved similarity plot at {}".format(figure_path))
print("saved regression plot at {}".format(reg_fig_path))