github.com/HyperGAN/HyperGAN commits | Ecosyste.ms: OpenCollective

[fix] sample save directory

8ccfa6fa855192219abd42f4d7abe1cf4781644f authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] optimistic loading when new layer is smaller

87b2a455c24ad6fd81016b6b79cb21392f484f2b authored over 6 years ago by Martyn Garcia <[email protected]>

[fix] missing ag1 variable

d4c5426a22d83c5d8e53f28b8a7f61c5e9f69342 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] crossing the curl approximation with manifold guided

ff7bc84a8d0375c5a36381fd468ad7075b893cfa authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] partial loading of weights on shape mismatch

9e0d81d998d28ec0306319137622e9c67fc31bea authored over 6 years ago by Martyn Garcia <[email protected]>

[fix] optimistic loading of smaller filter bank

ff2a5ac6eb609fd5aa9a35cd5841f26361f0d564 authored over 6 years ago by Martyn Garcia <[email protected]>

[fix] output loading file on curriculum trainer

fe8399c848cd26099aa8fbbcc749d07278f44130 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] temporary debug output

daf770ea3181860d0e078e1e0e0ad0c6fc247524 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] salvage weights on optimistic restore with changing sizes

dd8a946daf579c27b425bc0d23b6da8768303362 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] "slice 1 2 3" layer

e4f52b9482f664e47425d35adbdee4fb3d1bd4c2 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] optimistic loading shape mismatch output

4b09011528089db95852ad8dadd30dd7b8df0f44 authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] loss terms

9defc997c64487b32ee5079f7118b61d80c9228e authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] tweak loss terms

2a5fe7c443f25f323c53c60153231b4d17cbe652 authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] loss terms

ac9dace6bdfd5c7f380b4f973f020142eaab6704 authored over 6 years ago by Martyn Garcia <[email protected]>

[experiment] manifold guided test

32a9e5c986a03bab9319a26928b424b51d759857 authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] second discriminator

5731a35c9e2dd637c2d442893367330fc88cc48a authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] stale conv_layers reference

2a549e1415804db55151e106aa24796c0d81d542 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] curriculum trainer

58579976a05bc835084e163f4a47f12cc382e8b2 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] optionalize inherit

c2285cf2897a29f475a0078bb986e169fbbe84c4 authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] loss terms

4f0122e6e3d849c943895edf248c02cf12a25dfa authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] configs can inherit from other configs (limit one level)

0cdcd4990753d859494d267fdfe871d24a262310 authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] another next frame experiment

2b01e087e71bbef5e08786dc4e4df9acd50fdddd authored over 6 years ago by Martyn Garcia <[email protected]>

[experiment] manifold guided tweaks

9a9cd3e0d807c981230dbc09602fa92e6d62fbb4 authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] manifold guided

650595b0639e0c7a57178f0bc4268852066ea26a authored over 6 years ago by Martyn Garcia <[email protected]>

[experiment] next frame

7a5021b9e54df0485da1d5aaeeebe72f029d0833 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] experimental flag

ad70119089594ebc5ef7e158b805ce6f6882414d authored over 6 years ago by Martyn Garcia <[email protected]>

[fix] remove unused variable references

cc324a229f55105f9a2db79c242b2c48dc29d875 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] new next frame experiment

d57c7ad4bd987b51e9b0780bf6afb23422f9d1d8 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] prev_l2_loss config gate

077ad50e7354afa006c8d64b24435c1bf184faf0 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] infogan

9cb876add9a79f963375eb4f669ae0e12c1aeba8 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] next frame runtime errors

07c2d719988c607142e1b21e3fcce2682c1d0e1d authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] saving infogan

97fbdce559739fcbb93a36ceb863a95346230954 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] infogan on loss

e971b04c8dbc99e55cbb5adcd7672bc5efb69ab3 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] batch_walk works with batch size 1

8daefb7241b63dbbf32a810f89c1f5106b59c523 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] batch_walk works with batch size 1

c7320c3b31fb300e6bab95b5a2e29212f332f2db authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] name ops for graph modification

a2d8d59a07d43f5d54f7c83d20868909552587ca authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] optimistic restore to change saved graph https://github.com/tensorflow/tensorflow/issues/312

10fc7986bcef21a32794dd2dd389b7e96063b294 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] preload trained networks in gang

5bc65b80e16afd39f537d4a0b26bf55829716395 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] config gate prev_l2_loss

0124be5109b2a689e510b1d72e7ad0acf329cc56 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] gradient locally stable implementation

679d172866989b8b964190dc1d3a243259cd9df1 authored over 6 years ago by Mikkel Garcia <[email protected]>

[refactor] gan.g_vars and gan.d_vars

5ff7df6e307281aa859d2bed184d3f74124bb64e authored over 6 years ago by Martyn Garcia <[email protected]>

[refactor] remove invalid option

90d77f4d5a4dcd4c78af3b850007d83815e329e7 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] jacobian regularizer attempt https://arxiv.org/abs/1806.09235

904f2acaa04719bb710f213fe9c5e60866531c2b authored over 6 years ago by Mikkel Garcia <[email protected]>

[refactor] stop defaulting gan components

c50a5ffb91c32ea831f4c5b15ad2e9111d2cad74 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] manifold guided ali gan https://arxiv.org/pdf/1804.04391v1.pdf

b9f0248b132f69fc30f1308e6f31f02c9d04cf24 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] recreatable gan in gang step

b0bb1c13ee55eb2c9b2e179b7256303b61a56282 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] improving 2d distribution output

f822c8e799d8ebe05229124ca47c7c341213410e authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] gradients

c4177a31cc1ce4f9f193864661fa510193d0b5b2 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] initial attempt at implementing crossing-the-curl https://arxiv.org/pdf/1808.01531v1.pdf

b848249eb013e0e59fab655cb1d622b3790b71cf authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] optimize loss.random_penalty

6e95c7082692b80b7e0e6edeecc9b1e9c2902512 authored over 6 years ago by Martyn Garcia <[email protected]>

[feature] style_encoder for variational style

9aa2fdd8e21285274c49c4d48e1f54f283681052 authored over 6 years ago by Martyn Garcia <[email protected]>

[fix] use layer_conv instead of ops.conv2d

f9be77b23d946d78e0e07a07db33a7a3afb1621e authored over 6 years ago by Martyn Garcia <[email protected]>

[feature] grid sampler works with batch size 1

a4f9d20d37974248d3738e5c7d8a160d9cf0b2bb authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] syntax

d6bb42bc6c2eeb2709fffc671f51d0fef39b62af authored over 6 years ago by Martyn Garcia <[email protected]>

[feature] config.alpha

ccc92eef3f17751bb0c97cbd76a5ffc7c31e4abc authored over 6 years ago by Martyn Garcia <[email protected]>

[feature] variablize learn_rate

df7a329bb6c225bceaff9ba513d66806342e103c authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] orthogonal regularizer targets

cbdc37b193274e37b9a2fcce5095096d00ada4ee authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] orthogonal regularizer targets

c79808245dc516f08e8a2946509b9b97922e0be5 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] two sided activations and [fix] prelu reuse

cc0543dbdf9c51ebcae98b76ced4494d15dafd81 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] weight regularizers

7080b5b43882ca5bba584f329d67cf3d73cd996c authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] new penalty regularizers

9d7da45b67d36a711a7710c0d43d9eea792daa66 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] gan.weights()

258eb91cf21450e13e6eb898bc6894e85645ba26 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] 2 sided relu

034d0308f5294ee0d3f231409c2150e021a98a52 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] metrics setup for regularizers

6704942543b63c8d260624421c54c1f094bd42ea authored over 6 years ago by Martyn Garcia <[email protected]>

[fix] l2nn

c13442c95f2732156ce397c90f5108719e2c32b5 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] loading ema

e3fff8776cd7544e418a2aa8bef8574da99bafac authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] weight constraints

b6f8ed65ebec4e3afe9f287fab14136f203f18ff authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] two sided relu

31d544f80501ee05afb96071eb3c381dbe1526b8 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] l2_scaled

e87201c11009b6518797c368260b8b3ae698848b authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] skip_fitness option

30a9a216609b077829c028eb2bfb669b86e1dcc4 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] orthogonal weight constraint, lipschitz weight constraint

4f5e7c1e8963da92a310d56e4412a419c9889baf authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] exponential moving average on weights (ema)

35d9aa8a6d11ac5c30d0be81ed3b1be9ce503ec1 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] remove broken exponential moving average option

311461412b49ff7db0cac6babd9e9142ace8defc authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] metrics

905a9a24e53495199f7e517287501c008801f38c authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] random_penalty, as an alternative to k_lipchitz_penalty

1019c37b159f8dc9a9a058b6a039c2dbca607c07 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] lipschitz weight constraint

0ddbb8cdee836167faa65f161348bbedb3bf5f87 authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] k_lipschitz_ragan flag

06434f85dba08e1e10967918e9d55d6fb6e539aa authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] lipschitz weight constraint

7b0c574cb1ce55cc978e230a2864d44510edfb9b authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] attempt at lipschitz regularization https://arxiv.org/pdf/1804.04368.pdf

46df8bb521b266cf1e2dc53506728ebc6ab3b591 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] he_normal initializer

9387b3a1744a98d9d92e92391919d4d4e8c5a8a1 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] loss.k_lipschitz_penalty https://arxiv.org/pdf/1803.06107.pdf

d570be16036754eb409c1c4a25366de49512f650 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] pixel norm

847bb4ada74bfff8aac6fc522411f33d7d4dfe3d authored over 6 years ago by Martyn Garcia <[email protected]>

[feature] weight_constraint on fitness trainer

846d093e6fd4b2b0bf183d8bcf075fb8a3d41f3d authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] minibatch layer

9702ad353632d8178df88a1ef4f2b1b4c437fec3 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] crossover

e0ffc65bc4efc5a0346f21f487269c2958418d82 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] double rbbr create

3b6013727f98a5832fe784cf200bba661199c1eb authored over 6 years ago by Mikkel Garcia <[email protected]>

[refactor] print statements

117480959bfa7d280310d155e9ba0614678a6ee2 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] start g(z0) at random

8c4347a74e0651337b2d52d77e0613f6ea3a3368 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] prelu weight location

36e713ec5ec012558b88056596f465d1ead21e21 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] sample from uniform

5a961a3f17050880ac7c072325eaac5c0a7ee57e authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] gan.variables() returns all variables

21ff1572a349ebd2485cc46ae50b120884bb4789 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] ali terms

949d32434b52633de821df38510f91a7b8a8dbe3 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] remove stddev option

b4700794f78ed49f8848da12d1fdb5a116f81528 authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] remove stddev option

d20206902029adb51ec1529d00a0c6e4dfa27290 authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] sorted by d_fake sampler

fe06673fb7138e392452289192c8f1ca2e304b1f authored over 6 years ago by Mikkel Garcia <[email protected]>

[fix] l2 regularizer

6df7f960761d6d86dec11c0d1d279568f50d648a authored over 6 years ago by Martyn Garcia <[email protected]>

[feature] l2 loss on previous ||g(z)-g0(z)||2

8b61afd239ae70bdec7d2da538a09b305ec98e65 authored over 6 years ago by Mikkel Garcia <[email protected]>

[experiment] new fitness types

9d1e982e0d5968845f02994dc0823653f695e52c authored over 6 years ago by Mikkel Garcia <[email protected]>

[feature] gang trainer option mutate_child

6a7e13dfe74bf70e312fedec2d6afc7e7d381ae5 authored over 6 years ago by Martyn Garcia <[email protected]>

[feature] train_d_on_sgs

cf6b0e8378642f2b25a3b8cf8615fa135fb421d2 authored over 6 years ago by Mikkel Garcia <[email protected]>