Cleanup and optimize transformers #4348
Conversation
vinx13
commented
Jun 28, 2018
vinx13
commented
- Use linalg in NormOne
- Optimize PruneVarSubMean
vinx13
changed the title
vigsterkr
force-pushed
the
feature/transformers
branch
from
550639e to
af8af72
Compare
|
This needs a rebase |
|
Btw whenever you make something linalg, pls ensure there is some test coverage of the changes lines, so that we at least know he results didn’t change in the integration test examples |
vinx13
force-pushed
the
feature/transformers_update
branch
from
dedcb4a to
bdacb07
Compare
| EXPECT_EQ(v.vlen, num_features); | ||
| for (auto j : range(v.vlen)) | ||
| { | ||
| EXPECT_DOUBLE_EQ(v[j], data[num_features * i + j] / norm[i]); |
| { | ||
| SGVector<float64_t> v = feats->get_feature_vector(i); | ||
| auto result = transformer->apply_to_feature_vector(v); | ||
| EXPECT_EQ(result.vlen, num_features); |
| @@ -102,7 +102,7 @@ void CPruneVarSubMean::cleanup() | |||
| SGMatrix<float64_t> | |||
| CPruneVarSubMean::apply_to_matrix(SGMatrix<float64_t> matrix) | |||
| { | |||
| ASSERT(m_initialized) | |||
| REQUIRE(m_initialized, "Transformer has not been fitted.\n"); | |||
There was a problem hiding this comment.
offtopic: I would love to move all those general checks to base classes so that they dont get repeated in every specialization....
any thoughts on this
There was a problem hiding this comment.
shouldn't we use a different exception type other than ShogunException
vigsterkr
force-pushed
the
feature/transformers
branch
2 times, most recently
from
77a9f85 to
8711ea6
Compare
vinx13
force-pushed
the
feature/transformers_update
branch
from
c1b0299 to
47363cb
Compare
vinx13
force-pushed
the
feature/transformers_update
branch
from
47363cb to
918fead
Compare
| @@ -102,7 +102,7 @@ void CPruneVarSubMean::cleanup() | |||
| SGMatrix<float64_t> | |||
| CPruneVarSubMean::apply_to_matrix(SGMatrix<float64_t> matrix) | |||
| { | |||
| ASSERT(m_initialized) | |||
| REQUIRE(m_initialized, "Transformer has not been fitted.\n"); | |||
There was a problem hiding this comment.
shouldn't we use a different exception type other than ShogunException
| @@ -132,31 +132,18 @@ CPruneVarSubMean::apply_to_matrix(SGMatrix<float64_t> matrix) | |||
| /// result in feature matrix | |||
| SGVector<float64_t> CPruneVarSubMean::apply_to_feature_vector(SGVector<float64_t> vector) | |||
| { | |||
| float64_t* ret=NULL; | |||
| REQUIRE(m_initialized, "Transformer has not been fitted.\n"); | |||
There was a problem hiding this comment.
shouldn't we use a different exception type other than ShogunException
| MS_NOT_AVAILABLE); | ||
| } | ||
|
|
||
| void CTransformer::check_fitted() const |
There was a problem hiding this comment.
i would rather go with the java like: is_fitted or is_fit depending which english we'd like to go with :) but let's stick to british (non north american english as in case of the exception type naming)
| CTransformer::CTransformer() : CSGObject() | ||
| { | ||
| SG_ADD( | ||
| &m_fitted, "fitted", "Whether the transformer has been fitted.", |
There was a problem hiding this comment.
let's register this param as is_fitted
