N_out = 10
N_list = None
input_type = 'task_inp'
# output_type = 'factored'
# output_type = 'rotated1.0'
output_type = [0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]
task = util.RandomDichotomies(num_class, num_dich, overlap=ovlp)
sample_dichotomies = num_dich
# sample_dichotomies = None
this_exp = exp.random_patterns(task,
SAVE_DIR,
num_class=num_class,
dim=100,
var_means=1)
# FOLDERS = 'results/continuous/%d_%d/%s/%s/'%(num_class,num_dich,input_type, output_type)
# if (N_list is None):
# files = os.listdir(SAVE_DIR+FOLDERS)
# param_files = [f for f in files if 'parameters' in f]
# if len(param_files)==0:
# raise ValueError('No experiments in specified folder `^`')
# Ns = np.array([re.findall(r"_(\d+)_(\d+)?",f)[0] \
# for f in param_files]).astype(int)
this_exp = exp.mnist_multiclass(task, SAVE_DIR,
z_prior=latent_dist,
num_layer=num_layer,
weight_decay=decay,
decoder=readout_weights,
nonlinearity=nonlinearity,
good_start=good_start,
init_coding=coding_level)
elif which_task == 'mog':
task = util.RandomDichotomies(num_cond,num_var)
this_exp = exp.random_patterns(task, SAVE_DIR,
num_class=num_cond,
dim=100,
var_means=1,
z_prior=latent_dist,
num_layer=num_layer,
weight_decay=decay,
decoder=readout_weights,
nonlinearity=nonlinearity,
good_start=good_start,
init_coding=coding_level,
rot=rotation)
elif which_task == 'structured':
inp_task = tasks.EmbeddedCube(tasks.StandardBinary(int(np.log2(num_cond))),100,noise_var=0.1)
# inp_task = tasks.TwistedCube(tasks.StandardBinary(2), 100, f=rotation, noise_var=0.1)
# inp_task = tasks.NudgedXOR(tasks.StandardBinary(2), 100, nudge_mag=rotation, noise_var=0.1, random=True)
# task = tasks.LogicalFunctions(d=decs, function_class=num_var)
task = tasks.RandomDichotomies(d=[(0,1,3,5),(0,2,3,6),(0,1,2,4)])
# task = tasks.RandomDichotomies(d=[(0,3)])
this_exp = exp.structured_inputs(task, input_task=inp_task,
SAVE_DIR=SAVE_DIR,
noise_var=0.1,
import students
import assistants
import util
import experiments as exp
#%%
num_cond = 8
num_var = 2
task = util.RandomDichotomies(num_cond,num_var,0)
# task = util.ParityMagnitude()
# this_exp = exp.mnist_multiclass(task, SAVE_DIR, abstracts=abstract_variables)
this_exp = exp.random_patterns(task, SAVE_DIR,
num_class=8,
dim=100,
var_means=1,
var_noise=0.1)
#%% set up the task
p = 2**num_var
allowed_actions = [0,1,2]
# allowed_actions = [0]
p_action = [0.8,0.1,0.1]
# p_action = [1.0]
# output_states = this_exp.train_data[1].data
# output_states = util.decimal(this_exp.train_data[1])
output_states = this_exp.train_data[0].data
# output_states = ContinuousEmbedding(N_, 1.0)(this_exp.train_data[1])
elif this_task == 'mog':
task = tasks.RandomDichotomies(num_class,
num_dich,
overlap=ovlp,
use_mse=gaus_obs)
exp = experiments.random_patterns(N=N,
task=task,
SAVE_DIR=SAVE_DIR,
num_class=num_class,
dim=100,
var_means=1,
H=H,
nonlinearity=nonlinearity,
num_layer=num_layer,
z_prior=latent_dist,
weight_decay=decay,
decoder=readout_weights,
bsz=64,
lr=1e-3,
nepoch=nepoch,\
sample_dichotomies=sample_dichotomies,
init=init,
skip_metrics=skip_metrics,
init_coding=coding_level,
good_start=ols_initialised,
fix_decoder=fixed_decoder,
rot=rot)
elif this_task == 'structured':
# bits = np.nonzero(1-np.mod(np.arange(num_class)[:,None]//(2**np.arange(np.log2(num_class))[None,:]),2))
# pos_conds = np.split(bits[0][np.argsort(bits[1])],int(np.log2(num_class)))
# inp_task = tasks.EmbeddedCube(tasks.StandardBinary(int(np.log2(num_class))))
# inp_task = tasks.TwistedCube(tasks.StandardBinary(2), 100, f=coding_level, noise_var=0.1)
abstract_variables = util.DigitsBitwise()
# abstract_variables = util.ParityMagnitude()
# abstract_variables = util.RandomDichotomies(2)
# task = util.RandomDichotomies(8,2,0)
task = util.ParityMagnitude()
Q = abstract_variables.num_var
dic_type = 'general'
# dic_type = 'simple'
# this_exp = exp.mnist_multiclass(task, SAVE_DIR, abstracts=abstract_variables)
this_exp = exp.random_patterns(task,
SAVE_DIR,
num_class=8,
dim=100,
var_means=1,
abstracts=abstract_variables)
#%% make factorized representation
N = 100
C = np.random.rand(N, N)
W1 = la.qr(C)[0][:, :this_exp.dim_output]
targ = ((2 * this_exp.train_data[1].numpy() - 1) * 10) @ W1.T
b = 0.8 * targ.min()
linreg = linear_model.LinearRegression()
# linreg.fit(this_exp.train_data[0],this_exp.train_data[1].numpy()@W1.T)
linreg.fit(this_exp.train_data[0], targ - b)
Q = task.num_var
# N_list = None # set to None if you want to automatically discover which N have been tested
# N_list = [2,3,4,5,6,7,8,9,10,11,20,25,50,100]
# N_list = None
# N_list = [2,3,5,10,50,100]
N_list = [100]
# find experiments
# this_exp = exp.mnist_multiclass(task, SAVE_DIR,
# z_prior=latent_dist,
# num_layer=num_layer,
# weight_decay=decay)
this_exp = exp.random_patterns(task,
SAVE_DIR,
num_class=8,
dim=100,
var_means=1,
z_prior=latent_dist,
num_layer=num_layer,
weight_decay=decay)
this_folder = SAVE_DIR + this_exp.folder_hierarchy()
if (N_list is None):
files = os.listdir(this_folder)
param_files = [f for f in files if 'parameters' in f]
if len(param_files) == 0:
raise ValueError('No experiments in specified folder `^`')
Ns = np.array([re.findall(r"N(\d+)_%s"%nonlinearity,f)[0] \
for f in param_files]).astype(int)
