def plot_curve(model_class,
masking_frequency: float,
probe_level: int,
masking_level: int,
loss: float,
learned_vars: List[float],
save_path: str,
dataset,
other_axis,
all_losses=List[int],
baseline_losses=List[int]) -> str:
other_axis.set_title("Masker Frequency {} Hz, Probe Level {} dB".format(
masking_frequency, probe_level),
fontsize=12)
masking_level_to_color = {40: "#1f77b4", 60: "#aec7e8", 80: "#ff7f0e"}
# Get the data and model for this specs
data = dataset.get_curve_data(masking_frequency=masking_frequency,
probe_level=probe_level,
masking_level=masking_level)
actual_frequencies, actual_amplitudes = zip(*data)
highest_y = max(actual_amplitudes) + 1
all_losses.append(loss / len(actual_frequencies))
# Do inference with the model
predicted_amplitudes = []
for frequency in actual_frequencies:
current_inputs = [frequency] + learned_vars
predicted_amplitudes.append(model_class.function(*current_inputs))
# Sample 1000 points from the learned curve
sampled_frequencies = np.linspace(min(actual_frequencies),
max(actual_frequencies), 1000)
sampled_frequencies.sort()
sampled_amplitudes = []
for freq in sampled_frequencies:
current_inputs = [freq] + learned_vars
amps = model_class.function(*current_inputs)
sampled_amplitudes.append(amps)
# Plot everything individually and on the grid
error_lines = []
baseline_errors = []
for f, ampl_tuple in zip(actual_frequencies,
zip(actual_amplitudes, predicted_amplitudes)):
pair = [(f, ampl_tuple[0]), (f, ampl_tuple[1])]
error_lines.append(pair)
baseline_errors.append((0 - ampl_tuple[0])**2)
baseline_losses.append(sum(baseline_errors) / len(baseline_errors))
linecoll = matcoll.LineCollection(error_lines, colors="k")
linecoll2 = matcoll.LineCollection(error_lines, colors="k")
# Plot fitted data
cur_fig, ax = plt.subplots(num=2)
other_axis.axvline(x=model_class.masker_frequency_bark, color="C2", ls=":")
ax.axvline(model_class.masker_frequency_bark,
c="c",
label="Fixed Loc",
ls=":")
ax.scatter(actual_frequencies, actual_amplitudes, c="b", label="Actual")
other_axis.scatter(actual_frequencies,
actual_amplitudes,
label="Actual, Masker Level {}".format(masking_level))
ax.scatter(actual_frequencies,
predicted_amplitudes,
c="r",
label="Predicted")
ax.plot(sampled_frequencies, sampled_amplitudes)
other_axis.plot(sampled_frequencies,
sampled_amplitudes,
label="Learned, Masker Level {}".format(masking_level))
ax.add_collection(linecoll)
ax.legend()
other_axis.legend()
other_axis.add_collection(linecoll2)
plt.rcParams.update({"font.size": 8})
title = "Nelder-Mead \n Current Loss: %.2f, " % (loss / len(
actual_frequencies)) + model_class.parameter_repr(learned_vars)
ax.set_ylim(0, highest_y)
ax.set_title(title)
ax.set_xlim(0, 24)
other_axis.set_ylim(0, highest_y)
other_axis.set_xlim(0, 24)
filename = os.path.join(
save_path, "masker_{}_probe_level_{}_masking_level_{}.png".format(
masking_frequency, probe_level, masking_level))
plt.savefig(filename, dpi=cur_fig.dpi)
plt.close(cur_fig)
return filename, other_axis
flags.DEFINE_string("logs_dir", "logs", "Where the log file is saved.")
flags.DEFINE_string("logs_file", "logs.txt", "Where the logs are saved.")
flags.DEFINE_string("loss_binary", "./loss.py", "Binary to loss file")
flags.DEFINE_integer("plot_every_n_iterations", 10,
"How often to plot learning progress.")
dataset = dataset.MaskingDataset()
for filename in os.listdir("data"):
if filename.startswith("masker") and filename.endswith(".txt"):
dataset.read_data("data", filename)
masking_frequency = float(os.environ["MASK_FREQ"])
probe_level = int(os.environ["PROBE_LEVEL"])
masking_level = int(os.environ["MASKING_LEVEL"])
data = dataset.get_curve_data(masking_frequency=masking_frequency,
probe_level=probe_level,
masking_level=masking_level)
actual_frequencies, actual_amplitudes = zip(*data)
model_class = model.Model(masking_frequency, probe_level, masking_level)
def calculate_model_output(inputs: List[float], pars: List[float]) -> float:
model_vars = inputs + model_class.parameters_from_learned(pars)
output = model_class.function(*model_vars)
return output
def plot_simplex(current_vars: List[float], best_vars: List[float],
iteration: int, current_loss: float, best_loss: float,
highest_y: float, save_path: str) -> str:
def plot_curve(model_class, masking_frequency: float, probe_level: int, masker_frequency_bark: int,
masking_level: int,
learned_pars: List[float], save_path: str, dataset,
other_axis=None) -> str:
if other_axis:
other_axis.set_title("Masker Frequency {} Hz, Probe Level {} dB".format(
masking_frequency, probe_level), fontsize=12)
masking_level_to_color = {40: "#1f77b4", 60: "#aec7e8", 80: "#ff7f0e"}
# Get the data and model for this specs
try:
data = dataset.get_curve_data(
masking_frequency=masking_frequency,
probe_level=probe_level,
masking_level=masking_level)
except ValueError:
return "", None
actual_frequencies, actual_amplitudes = zip(*data)
highest_y = max(actual_amplitudes) + 1
# Do inference with the model
predicted_amplitudes = []
for frequency in actual_frequencies:
prediction = model_class.predict(masking_frequency, probe_level,
masking_level, frequency)
predicted_amplitudes.append(prediction)
# Sample 1000 points from the learned curve
sampled_frequencies = list(range(0, 25))
sampled_frequencies.sort()
sampled_amplitudes = []
for freq in sampled_frequencies:
prediction = model_class.predict(masking_frequency, probe_level,
masking_level, freq)
sampled_amplitudes.append(prediction)
# Plot everything individually and on the grid
error_lines = []
for f, ampl_tuple in zip(actual_frequencies,
zip(actual_amplitudes, predicted_amplitudes)):
pair = [(f, ampl_tuple[0]), (f, ampl_tuple[1])]
error_lines.append(pair)
linecoll = matcoll.LineCollection(error_lines, colors="k")
linecoll2 = matcoll.LineCollection(error_lines, colors="k")
# Plot fitted data
cur_fig, ax = plt.subplots(num=2)
if other_axis:
other_axis.axvline(x=masker_frequency_bark, color="C2", ls=":")
other_axis.scatter(
actual_frequencies,
actual_amplitudes,
label="Actual, Masker Level {}".format(masking_level))
other_axis.plot(
sampled_frequencies,
sampled_amplitudes,
label="Learned, Masker Level {}".format(masking_level), ls="--")
other_axis.legend()
other_axis.add_collection(linecoll2)
other_axis.set_ylim(0, highest_y)
other_axis.set_xlim(0, 24)
ax.axvline(masker_frequency_bark, c="c", label="Fixed Loc", ls=":")
ax.scatter(actual_frequencies, actual_amplitudes, c="b", label="Actual")
ax.scatter(actual_frequencies, predicted_amplitudes, c="r", label="Predicted",
ls="--")
ax.plot(sampled_frequencies, sampled_amplitudes)
ax.add_collection(linecoll)
ax.legend()
plt.rcParams.update({"font.size": 8})
title = "Predictions on Test Set"
ax.set_ylim(0, highest_y)
ax.set_xlim(0, 24)
ax.set_title(title)
filename = os.path.join(
save_path, "masker_{}_probe_level_{}_masking_level_{}.png".format(
masking_frequency, probe_level, masking_level))
plt.savefig(filename, dpi=cur_fig.dpi)
plt.close(cur_fig)
return filename, other_axis