diff --git a/deepof/models.py b/deepof/models.py
index 3ce3e5b6a62a383cc010f8646050b5e6fd1cd3c8..29e73854e5aa93cccc49912ed48c58cb4388664a 100644
--- a/deepof/models.py
+++ b/deepof/models.py
@@ -624,7 +624,7 @@ class SEQ_2_SEQ_GMVAE:
name="encoding_distribution",
)([z_cat, z_gauss])
- encode_to_distribution = Model(x, z, name="SEQ_2_SEQ_trained_distribution")
+ posterior = Model(x, z, name="SEQ_2_SEQ_trained_distribution")
# Define and control custom loss functions
if "ELBO" in self.loss:
@@ -679,7 +679,7 @@ class SEQ_2_SEQ_GMVAE:
)(generator)
# define individual branches as models
- encode_to_vector = Model(x, z, name="SEQ_2_SEQ_VEncoder")
+ encoder = Model(x, z, name="SEQ_2_SEQ_VEncoder")
generator = Model(g, x_decoded_mean, name="vae_reconstruction")
def log_loss(x_true, p_x_q_given_z):
@@ -687,7 +687,7 @@ class SEQ_2_SEQ_GMVAE:
the output distribution"""
return -tf.reduce_sum(p_x_q_given_z.log_prob(x_true))
- model_outs = [generator(encode_to_vector.outputs)]
+ model_outs = [generator(encoder.outputs)]
model_losses = [log_loss]
model_metrics = {"vae_reconstruction": ["mae", "mse"]}
loss_weights = [1.0]
@@ -736,9 +736,9 @@ class SEQ_2_SEQ_GMVAE:
loss_weights.append(self.phenotype_prediction)
# define grouper and end-to-end autoencoder model
- grouper = Model(encode_to_vector.inputs, z_cat, name="Deep_Gaussian_Mixture_clustering")
+ grouper = Model(encoder.inputs, z_cat, name="Deep_Gaussian_Mixture_clustering")
gmvaep = Model(
- inputs=encode_to_vector.inputs,
+ inputs=encoder.inputs,
outputs=model_outs,
name="SEQ_2_SEQ_GMVAE",
)
@@ -754,11 +754,12 @@ class SEQ_2_SEQ_GMVAE:
gmvaep.build(input_shape)
return (
- encode_to_vector,
- encode_to_distribution,
+ encoder,
generator,
grouper,
gmvaep,
+ self.prior,
+ posterior,
)
@prior.setter
diff --git a/supplementary_notebooks/deepof_model_evaluation.ipynb b/supplementary_notebooks/deepof_model_evaluation.ipynb
index de62ca5557c95aa9440697eb9cb3304a7a28dea3..8e1bbe4075a4b6afa2a5190d43aab72a0b75de46 100644
--- a/supplementary_notebooks/deepof_model_evaluation.ipynb
+++ b/supplementary_notebooks/deepof_model_evaluation.ipynb
@@ -15219,165 +15219,24 @@
},
{
"cell_type": "code",
- "execution_count": 18,
- "metadata": {},
- "outputs": [],
- "source": [
- "# Retrieve latent distribution parameters and sample from posterior\n",
- "def get_median_params(component, categories, cluster, param):\n",
- " # means = [np.median(component.mean().numpy(), axis=0) for component in mix_components]\n",
- " # stddevs = [np.median(component.stddev().numpy(), axis=0) for component in mix_components]\n",
- " if param == \"mean\":\n",
- " component = component.mean().numpy()\n",
- " elif param == \"stddev\":\n",
- " component = component.stddev().numpy()\n",
- "\n",
- " cluster_select = np.argmax(categories, axis=1) == cluster\n",
- " if np.sum(cluster_select) == 0:\n",
- " return None\n",
- " component = component[cluster_select]\n",
- " return np.median(component, axis=0)"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 19,
- "metadata": {},
- "outputs": [],
- "source": [
- "def retrieve_latent_parameters(\n",
- " distribution, reduce=False, plot=False, categories=None, filt=0, save=True\n",
- "):\n",
- " mix_components = distribution.components\n",
- "\n",
- " # The main problem is here! We need to select only those training instances in which a given cluster was selected.\n",
- " # Then compute the median for those only\n",
- "\n",
- " means = [\n",
- " get_median_params(component, categories, i, \"mean\")\n",
- " for i, component in enumerate(mix_components)\n",
- " ]\n",
- " stddevs = [\n",
- " get_median_params(component, categories, i, \"stddev\")\n",
- " for i, component in enumerate(mix_components)\n",
- " ]\n",
- " means = [i for i in means if i is not None]\n",
- " stddevs = [i for i in stddevs if i is not None]\n",
- "\n",
- " filts = np.max(categories, axis=0) > filt\n",
- " means = [i for i, j in zip(means, filts) if j]\n",
- " stddevs = [i for i, j in zip(stddevs, filts) if j]\n",
- "\n",
- " return means, stddevs\n",
- "\n",
- "\n",
- "def sample_from_posterior(\n",
- " decoder, parameters, component, enable_variance=False, samples=1\n",
- "):\n",
- " means, stddevs = parameters\n",
- " sample = np.random.normal(\n",
- " size=[samples, len(means[component])],\n",
- " loc=means[component],\n",
- " scale=(stddevs[component] if enable_variance else 0),\n",
- " )\n",
- " reconstruction = decoder(sample).mean()\n",
- "\n",
- " return reconstruction"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 105,
- "metadata": {},
- "outputs": [],
- "source": [
- "samples = np.random.choice(range(data_prep.shape[0]), 10000)\n",
- "latent_distribution = encode_to_distribution(data_prep[samples])"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 101,
- "metadata": {},
- "outputs": [],
- "source": [
- "means, stddevs = retrieve_latent_parameters(\n",
- " latent_distribution,\n",
- " categories=groupings[samples],\n",
- " reduce=False,\n",
- " plot=False,\n",
- " filt=0.,\n",
- " save=False,\n",
- ")"
- ]
- },
- {
- "cell_type": "code",
- "execution_count": 104,
+ "execution_count": 153,
"metadata": {},
"outputs": [
{
- "data": {
- "application/vnd.jupyter.widget-view+json": {
- "model_id": "fd7e515df218432b88e6fa7a6a75f9de",
- "version_major": 2,
- "version_minor": 0
- },
- "text/plain": [
- "interactive(children=(Dropdown(description='dim_red', options=('PCA', 'LDA', 'umap', 'tSNE'), value='PCA'), In…"
- ]
- },
- "metadata": {},
- "output_type": "display_data"
+ "ename": "AttributeError",
+ "evalue": "'Functional' object has no attribute 'prior'",
+ "output_type": "error",
+ "traceback": [
+ "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+ "\u001b[0;31mAttributeError\u001b[0m Traceback (most recent call last)",
+ "\u001b[0;32m<ipython-input-153-0d3135aace65>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m 1\u001b[0m \u001b[0;31m# Get prior distribution\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mgmvaep\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mprior\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
+ "\u001b[0;31mAttributeError\u001b[0m: 'Functional' object has no attribute 'prior'"
+ ]
}
],
"source": [
- "# Plot sampled data in the latent space\n",
- "\n",
- "\n",
- "@interact(dim_red=[\"PCA\", \"LDA\", \"umap\", \"tSNE\"], samples=(5000, 15000))\n",
- "def plot_static_latent_space(dim_red, samples):\n",
- " if dim_red == \"umap\":\n",
- " reducer = umap.UMAP(n_components=2)\n",
- " elif dim_red == \"LDA\":\n",
- " reducer = LinearDiscriminantAnalysis(n_components=2)\n",
- " elif dim_red == \"PCA\":\n",
- " reducer = PCA(n_components=2)\n",
- " else:\n",
- " reducer = TSNE(n_components=2)\n",
- "\n",
- " categories = latent_distribution.cat.sample(samples).numpy().flatten()\n",
- " mixture_sample = np.squeeze(\n",
- " np.concatenate(\n",
- " [latent_distribution.components[i].sample(1) for i in categories]\n",
- " )\n",
- " )\n",
- "\n",
- " print(mixture_sample.shape)\n",
- "\n",
- " if dim_red != \"LDA\":\n",
- " enc = reducer.fit_transform(mixture_sample)\n",
- " else:\n",
- " enc = reducer.fit_transform(\n",
- " mixture_sample,\n",
- " np.repeat(range(len(latent_distribution.components)), categories),\n",
- " )\n",
- "\n",
- " plt.figure(figsize=(12, 8))\n",
- "\n",
- " sns.scatterplot(enc[:, 0], enc[:, 1], hue=categories, palette=\"muted\")\n",
- " plt.title(\n",
- " \"Mean representation of latent space - K={}/{} - L={}\".format(\n",
- " len(means),\n",
- " len(latent_distribution.components),\n",
- " len(latent_distribution.components),\n",
- " )\n",
- " )\n",
- "\n",
- " plt.xlabel(\"{} 1\".format(dim_red))\n",
- " plt.ylabel(\"{} 2\".format(dim_red))\n",
- " plt.suptitle(\"Static view of trained latent space\")\n",
- " plt.show()"
+ "# Get prior distribution\n",
+ "gmvaep.prior"
]
},
{