def createPolyFit(coefs):
# create partical function
partialFit = part(polyFit, coefs)
return partialFit
def __init__(self,
vocabulary_hash,
vocabulary_size,
embedding_size,
sentence_size, memory_size, hops=3,
weight_tying_scheme=ADJACENT):
# Model Configurations.
super().__init__()
self.vocabulary_hash = vocabulary_hash
self.vocabulary_size = vocabulary_size
self.embedding_size = embedding_size
self.sentence_size = sentence_size
self.memory_size = memory_size
self.hops = hops
self.weight_tying_scheme = weight_tying_scheme
# Validate Configurations.
assert self.weight_tying_scheme in self.WEIGHT_TYING_SCHEMES, (
'Available weight tying schemes are: {schemes}'
.format(schemes=self.WEIGHT_TYING_SCHEMES)
)
# Memories.
self.A_memories = nn.ModuleList()
self.C_memories = nn.ModuleList()
for i in range(hops):
# Check if there's any previous memory layer.
y = i > 0
A_prev = self.A_memories[i-1] if y else None
C_prev = self.C_memories[i-1] if y else None
# 2.2.1. Adjacent Weight Tying
if self.tied_adjacent:
A_embedding = C_prev.embedding if y else None
C_embedding = None
A_temporal_embedding = C_prev.temporal_embedding if y else None
C_temporal_embedding = None
# 2.2.2. Layer-Wise Weight Tying
elif self.tied_layers:
A_embedding = A_prev.embedding if y else None
C_embedding = C_prev.embedding if y else None
A_temporal_embedding = A_prev.temporal_embedding if y else None
C_temporal_embedding = C_prev.temporal_embedding if y else None
# No Weight Tying
else:
A_embedding = None
C_embedding = None
A_temporal_embedding = None
C_temporal_embedding = None
self.A_memories.append(Memory(
self.vocabulary_size,
self.embedding_size,
self.sentence_size,
self.memory_size,
embedding=A_embedding,
temporal_embedding=A_temporal_embedding,
))
self.C_memories.append(Memory(
self.vocabulary_size,
self.embedding_size,
self.sentence_size,
self.memory_size,
embedding=C_embedding,
temporal_embedding=C_temporal_embedding
))
# Affine layer & Query embedding layer.
if self.tied_adjacent:
deepest_embedding = self.C_memories[-1].embedding
first_embedding = self.A_memories[0].embedding
self.linear = part(F.linear, weight=deepest_embedding.weight)
self.query_embedding = first_embedding
else:
self.linear = nn.Linear(self.embedding_size, self.vocabulary_size)
self.query_embedding = nn.Embedding(
self.vocabulary_size,
self.embedding_size
)
def simple_filter(signal: Signal, cutoff_freq: float, type: str) -> Signal:
# cutoff_freq - Hz
filtered_data = simple_filter_raw(signal.data, cutoff_freq,
signal.frequency, type)
return Signal(signal.info, filtered_data)
# TODO: pretty sure the amplitude changes after filtering,
# so the new info could be incorrect
lowpass_filter_sig = ['cutoff_freq']
lowpass_default_params = m(cutoff_freq=10 * Hz)
# lowpass_filter : (Signal, cutoff_freq: float) -> float
lowpass_filter = part(simple_filter, type='low')
lowpass_tr = Trans(
'Lowpass filter',
lowpass_filter,
lowpass_filter_sig,
)
highpass_filter_sig = ['cutoff_freq']
highpass_default_params = m(cutoff_freq=0.05 * Hz)
# highpass_filter : (Signal, float) -> float
highpass_filter = part(simple_filter, type='high')
highpass_tr = Trans('Highpass filter', highpass_filter, highpass_filter_sig)
######################################################################
# IMPORTANT - DON'T DELETE THESE CLASSES WITHOUT READING THE COMMENT #
######################################################################
eff_res[ACTIONS]
) # so we can process actions emitted during updating, if any
for command in eff_res[EFFECTS]:
state, eff_res = run_eff(handle(state, command),
signal_id=current_signal_id)
current_signal_id = eff_res[SIGNAL_ID]
return state
from debug_util import varied_dict_to_str as dts
s = lambda f, g: lambda x: g(f(x))
pd = ch(dts, print)
np = ch(part(map, str), part(str.join, '\n'), print)
nps = ch(sorted, np)
from typing import NamedTuple
from collections import OrderedDict
# from sumtype import sumtype
NodeId = NamedTuple('NodeId', [('id_', int)])
InputSlotId = NamedTuple('InputSlotId', [('node_id', NodeId), ('ix', int)])
OutputSlotId = NamedTuple('OutputSlotId', [('node_id', NodeId), ('ix', int)])
node_names = {NodeId(0): 'x', NodeId(1): 'y', NodeId(2): 'z'}
node_slot_ns = {NodeId(0): (1, 2), NodeId(1): (2, 0), NodeId(2): (0, 1)}
