def plot_2Dquiver_paths(self,
session,
Xdata,
Xvar_name='X:0',
scope="",
rlt_dir=TEST_DIR + addDateTime() + '/',
rslt_file='quiver_plot',
with_inflow=False,
savefig=False,
draw=False,
pause=False,
skipped=1,
feed_range=True,
range_xs=20.0,
Id=0):
"""
Plots a superposition of the 2D quiver plot and the paths in latent
space. Useful to check that the trajectories roughly follow the
dynamics.
"""
if savefig:
if not os.path.exists(rlt_dir): os.makedirs(rlt_dir)
rslt_file = rlt_dir + rslt_file
import matplotlib.pyplot as plt
axes = self.plot2D_sampleX(Xdata,
pause=pause,
draw=draw,
newfig=True,
skipped=skipped)
if feed_range:
x1range, x2range = axes.get_xlim(), axes.get_ylim()
s = int(5 * max(
abs(x1range[0]) + abs(x1range[1]),
abs(x2range[0]) + abs(x2range[1])) / 3)
else:
x1range = x2range = (-range_xs, range_xs)
s = int(5 * max(
abs(x1range[0]) + abs(x1range[1]),
abs(x2range[0]) + abs(x2range[1])) / 3)
self.quiver2D_flow(session,
Xvar_name=Xvar_name,
scope=scope,
pause=pause,
x1range=x1range,
x2range=x2range,
scale=s,
newfig=False,
with_inflow=with_inflow,
draw=draw,
Id=Id)
if savefig:
plt.savefig(rslt_file)
else:
pass
plt.close()
def create_RLT_DIR(Experiment_params):
# create the dir to save data
# Experiment_params is a dict containing param_name¶m pair
# Experiment_params must contain "rslt_dir_name":rslt_dir_name
cur_date = addDateTime()
local_rlt_root = './rslts/' + Experiment_params['rslt_dir_name'] + '/'
params_str = ""
for param_name, param in Experiment_params.items():
if param_name == 'rslt_dir_name':
continue
params_str += param_name + '_' + str(param) + '_'
RLT_DIR = local_rlt_root + params_str + cur_date + '/'
if not os.path.exists(RLT_DIR): os.makedirs(RLT_DIR)
return RLT_DIR
def plot_2Dquiver_paths(self,
session,
Xdata,
Xvar_name,
rlt_dir=TEST_DIR + addDateTime() + '/',
rslt_file='quiver_plot',
with_inflow=False,
savefig=False,
draw=False,
pause=False):
"""
"""
if savefig:
if not os.path.exists(rlt_dir): os.makedirs(rlt_dir)
rslt_file = rlt_dir + rslt_file
import matplotlib.pyplot as plt
axes = self.plot2D_sampleX(Xdata, pause=pause, draw=draw, newfig=True)
x1range, x2range = axes.get_xlim(), axes.get_ylim()
s = int(5 * max(
abs(x1range[0]) + abs(x1range[1]),
abs(x2range[0]) + abs(x2range[1])) / 3)
self.quiver2D_flow(session,
Xvar_name,
pause=pause,
x1range=x1range,
x2range=x2range,
scale=s,
newfig=False,
with_inflow=with_inflow,
draw=draw)
if savefig:
plt.savefig(rslt_file)
else:
pass
plt.close()
import sys import os from optparse import OptionParser import cPickle as pickle import numpy as np import theano import lasagne from lasagne.nonlinearities import softmax, linear, softplus # Local imports from LatEvModels import LocallyLinearEvolution from ObservationModels import GaussianObsTSGM, PoissonObsTSGM from datetools import addDateTime cur_date = addDateTime() # ============================================================================== LOCAL_RLT_ROOT = "/Users/danielhernandez/Work/time_series/vae_nlds_rec_algo_v2/data/" RLT_DIR = "poisson_data_008/" LOAD_DIR = LOCAL_RLT_ROOT + RLT_DIR # "/Users/danielhernandez/Work/time_series/vae_nlds_rec_algo_v2/data/gaussian_data_002/" LOAD_FILE = "datadict" YDIM = 10 XDIM = 2 TBINS = 30 NSAMPS = 100 OBSERVATIONS = 'Poisson' NNODES = 60 ALPHA = 0.0