def fit_curve(file_name, num_cdr3s, initial_param_guesses):
# get data
sample = data_utils.get_cdr3_counter_from_file(file_name, file_name)
# get x and y
x, y = sample.get_x_y_floats(num_cdr3s)
# define function to fit
def f(x, s, v_shift_left, piece_boundary, a, b, v_shift_right):
return np.piecewise(
x,
[
x < piece_boundary,
x >= piece_boundary,
],
[
lambda x: (x**-s) / zetac(s) + v_shift_left,
lambda x: a * x**b + v_shift_right,
],
)
# fit curve
fitted_params = curve_fit(f, x, y, p0=initial_param_guesses)[0]
print('fitted params:', fitted_params)
# measure error
score = r2_score(y_true=y, y_pred=f(x, *fitted_params))
print('R^2 score:', score)
# plot data and curve
x_dense = np.linspace(1, len(x), 100)
y_dense = f(x_dense, *fitted_params)
plot.scatter_and_func(x, y, x_dense, y_dense)
def find_common_cdr3s(file_names, n=None):
''' user facing function '''
samples = [data_utils.get_cdr3_counter_from_file(f, f) for f in file_names]
cdr3s = get_common_cdr3s(
samples,
limit=n,
)
print(cdr3s)
def cluster_cdr3s(file_name, dist_func, n_gram_len=1, n=25):
''' user facing function '''
sample = data_utils.get_cdr3_counter_from_file('s', file_name)
cdr3s = sample.get_sorted_cdr3s(limit=n)
ngrammed_dist_func = on_n_gram(n_gram_len)(dist_func)
hierarchical_clustering_cdr3s(
cdr3s=cdr3s,
dist_func=ngrammed_dist_func,
)
def get_n_closest_cdr3s(file_name, cdr3, n, dist_func, n_gram_len=1):
''' user facing function '''
sample = data_utils.get_cdr3_counter_from_file('s', file_name)
ngrammed_dist_func = on_n_gram(n_gram_len)(dist_func)
closest_cdr3s = get_closest_cdr3s_with_frequency(
cdr3=cdr3,
sample=sample,
dist_func=ngrammed_dist_func,
limit=n,
)
pprint(closest_cdr3s)
def preprocess_X(X, cdr3_limit):
''' preprocess X '''
# convert file names to sample vectors
counters = [data_utils.get_cdr3_counter_from_file(f, f) for f in X]
# top n CDR3s in samples
topc_counters = [c.get_sorted_sample(limit=cdr3_limit) for c in counters]
# weak intersection of CDR3s in samples
trimmed_counters = Sample.weak_intersection(topc_counters)
# convert to a dictionary that is compatible with a Pandas DataFrame
X = {i: c for i, c in enumerate(trimmed_counters)}
return X
def calculate_one(cdr3s, file_names, show_legend=True):
'''
Calculate a metric on a single distance.
'''
# get input data
counters = [
data_utils.get_cdr3_counter_from_file(f, f) for f in file_names
]
# calculate
dates = [data_utils.get_date_from_file_name(f) for f in file_names]
x, ys = calculate_lifespan(cdr3s, dates, counters)
# output results
short_log.info(f'cdr3s={cdr3s}, x={x}, ys={ys}, fnames={file_names}')
plot.lifespan_graph(cdr3s, x, ys, show_legend)
def main():
s = data_utils.get_cdr3_counter_from_file(
's', 'cdr3.b.A_2019_2020_d_00_20857.ann')
calculate_one(
cdr3s=[s.get_cdr3_by_rank(r) for r in range(1, 5)],
file_names=to_beta([
'cdr3.a.A_2017_2018_d_00_53535.ann',
'cdr3.a.A_2017_2018_d_07_11143.ann',
'cdr3.a.A_2017_2018_d_28_44887.ann',
'cdr3.a.A_2017_2018_m_04_73516.ann',
'cdr3.a.A_2019_2020_d_00_20857.ann',
]),
show_legend=True,
)
return 'done'
def main():
# get data
f = FILE_NAME
sample = data_utils.get_cdr3_counter_from_file(f,f)
# get x and y
x,y = sample.get_x_y(20)
# define function to fit
def f(x, a, v_shift):
return (x**-a)/zetac(a) + v_shift
# fit curve
fitted_params = curve_fit(f, x, y, p0=[6.1, min(y)])[0]
# plot data and curve
x_dense = np.linspace(1, len(x), 100)
y_dense = f(x_dense, *fitted_params)
plot.scatter_and_func(x, y, x_dense, y_dense)
def main():
# get data
f = 'cdr3.a.A_2017_2018_d_00_53535.ann'
sample = data_utils.get_cdr3_counter_from_file(f, f)
# get x and y
x, y = sample.get_x_y(500)
# define function to fit
def f(x, a, b, c):
return a * x**b + c
# fit curve
fitted_params = curve_fit(f, x, y, p0=[60000, -1.1, min(y)])[0]
print('fitted params:', fitted_params)
# measure error
score = r2_score(y_true=y, y_pred=f(x, *fitted_params))
print('R^2 score:', score)
# plot data and curve
x_dense = np.linspace(1, len(x), 100)
y_dense = f(x_dense, *fitted_params)
plot.scatter_and_func(x, y, x_dense, y_dense)