Ken Furudate

In [ ]:
import scanpy as sc
import anndata as ad
import squidpy as sq

import numpy as np
import pandas as pd

import matplotlib as mpl
from matplotlib import pyplot as plt
%matplotlib inline
import matplotlib.font_manager
plt.rcParams['font.sans-serif'] = ['Arial'] + plt.rcParams['font.sans-serif']
plt.rcParams["font.size"] = 20

import os
import seaborn as sns
from pathlib import Path

from IPython.core.display import display, HTML
display(HTML("<style>.container { width:80% !important; }</style>"))

import warnings
warnings.filterwarnings('ignore')

sc.logging.print_header()
print(f"squidpy=={sq.__version__}")

from anndata import AnnData

import skimage
import tangram as tg

/usr/local/lib/python3.7/dist-packages/distributed/config.py:20: YAMLLoadWarning: calling yaml.load() without Loader=... is deprecated, as the default Loader is unsafe. Please read https://msg.pyyaml.org/load for full details.
  defaults = yaml.load(f)

scanpy==1.9.1 anndata==0.8.0 umap==0.5.3 numpy==1.21.6 scipy==1.4.1 pandas==1.3.5 scikit-learn==1.0.2 statsmodels==0.13.2 python-igraph==0.9.11 pynndescent==0.5.7
squidpy==1.1.2
In [ ]:
datadir="/data/spatial/"
In [ ]:
in_f = "integrated_data.h5ad"

data = sc.read_h5ad(datadir + in_f)
data
Out[ ]:
AnnData object with n_obs × n_vars = 3637 × 4000
    obs: 'in_tissue', 'array_row', 'array_col', 'imagecol', 'imagerow', 'pathology', 'category', 'cluster', 'sample', 'n_counts', 'batch', '_scvi_batch', '_scvi_labels', 'leiden', 'sample_density', '_scvi_raw_norm_scaling'
    var: 'gene_ids', 'feature_types', 'genome', 'n_cells-0-0', 'n_cells-1-0', 'n_cells-1', 'highly_variable', 'highly_variable_rank', 'means', 'variances', 'variances_norm', 'highly_variable_nbatches'
    uns: '_scvi', 'dendrogram_leiden', 'hvg', 'leiden', 'leiden_colors', 'neighbors', 'pathology_colors', 'sample_colors', 'sample_density_params', 'umap', 'wilcoxon'
    obsm: 'X_pca', 'X_scvi', 'X_umap', 'spatial'
    layers: 'counts', 'scvi_expr'
    obsp: 'connectivities', 'distances'
In [ ]:
in_f1 = "SampleA_cluster_patho.annot_region.txt"
in_f2 = "SampleB_cluster_patho.annot_region.txt"
in_f3 = "SampleC_cluster_patho.annot_region.txt"
in_f4 = "SampleD_cluster_patho.annot_region.txt"
In [ ]:
a_annot = pd.read_table(os.path.join(datadir, "patho_annot", in_f1), index_col="Barcode")
b_annot = pd.read_table(os.path.join(datadir, "patho_annot", in_f2), index_col="Barcode")
c_annot = pd.read_table(os.path.join(datadir, "patho_annot", in_f3), index_col="Barcode")
d_annot = pd.read_table(os.path.join(datadir, "patho_annot", in_f4), index_col="Barcode")
In [ ]:
adata.obs['pathology'] = a_annot["patho_diag"]
adata.obs['category'] = a_annot["category"]
adata.obs['cluster'] = a_annot["graph-based"]

bdata.obs['pathology'] = b_annot["patho_diag"]
bdata.obs['category'] = b_annot["category"]
bdata.obs['cluster'] = b_annot["graph-based"]

cdata.obs['pathology'] = c_annot["patho_diag"]
cdata.obs['category'] = c_annot["category"]
cdata.obs['cluster'] = c_annot["graph-based"]

ddata.obs['pathology'] = d_annot["patho_diag"]
ddata.obs['category'] = d_annot["category"]
ddata.obs['cluster'] = d_annot["graph-based"]
In [ ]:
sample_name1="A"
sample_name2="B"
sample_name3="C"
sample_name4="D"
In [ ]:
adata.obs['sample'] = sample_name1
bdata.obs['sample'] = sample_name2
cdata.obs['sample'] = sample_name3
ddata.obs['sample'] = sample_name4
In [ ]:
adata_tumor = adata[adata.obs['category']=="tumor"]
bdata_tumor = bdata[bdata.obs['category']=="tumor"]
cdata_tumor = cdata[cdata.obs['category']=="tumor"]
ddata_tumor = ddata[ddata.obs['category']=="tumor"]
In [ ]:
sc.pp.filter_cells(adata_tumor, min_counts=500)
sc.pp.filter_cells(bdata_tumor, min_counts=500)
sc.pp.filter_cells(cdata_tumor, min_counts=500)
sc.pp.filter_cells(ddata_tumor, min_counts=500)

sc.pp.filter_genes(adata_tumor, min_cells=2)
sc.pp.filter_genes(bdata_tumor, min_cells=2)
sc.pp.filter_genes(cdata_tumor, min_cells=2)
sc.pp.filter_genes(ddata_tumor, min_cells=2)

filtered out 4 cells that have less than 500 counts
In [ ]:
def set_param(input_data, sample):
  scale = input_data.uns['spatial'][f"{sample}"]['scalefactors']['tissue_hires_scalef']
  img = sq.im.ImageContainer(input_data.uns['spatial'][f"{sample}"]['images']['hires'],
                           scale=scale, 
                           library_id=f"{sample}")
  img.show()
  return scale, img
In [ ]:
scale_a, img_a = set_param(adata_tumor, sample_name1)
scale_b, img_b = set_param(bdata_tumor, sample_name2)
scale_c, img_c = set_param(cdata_tumor, sample_name3)
scale_d, img_d = set_param(ddata_tumor, sample_name4)

Adding image layer `image`
Adding image layer `image`
Adding image layer `image`
Adding image layer `image`
In [ ]:
fig, axes = plt.subplots(1, 4, figsize=(20, 4))
img_a.show("image", cmap="gray", ax=axes[0])
axes[1].imshow(img_a["image"][:, :, 0, 0])
_ = sns.histplot(np.array(img_b["image"]).flatten(), ax=axes[2])
axes[3].imshow(img_a["image"][:, :, 0, 0]<0.6)
plt.tight_layout()
In [ ]:
fig, axes = plt.subplots(1, 4, figsize=(20, 4))
img_b.show("image", cmap="gray", ax=axes[0])
axes[1].imshow(img_b["image"][:, :, 0, 0])
_ = sns.histplot(np.array(img_b["image"]).flatten(), ax=axes[2])
axes[3].imshow(img_b["image"][:, :, 0, 0]<0.58)
plt.tight_layout()
In [ ]:
fig, axes = plt.subplots(1, 4, figsize=(20, 4))
img_c.show("image", cmap="gray", ax=axes[0])
axes[1].imshow(img_c["image"][:, :, 0, 0])
_ = sns.histplot(np.array(img_c["image"]).flatten(), ax=axes[2])
axes[3].imshow(img_c["image"][:, :, 0, 0]<0.78)
plt.tight_layout()
In [ ]:
fig, axes = plt.subplots(1, 4, figsize=(20, 4))
img_d.show("image", cmap="gray", ax=axes[0])
axes[1].imshow(img_d["image"][:, :, 0, 0])
_ = sns.histplot(np.array(img_d["image"]).flatten(), ax=axes[2])
axes[3].imshow(img_d["image"][:, :, 0, 0]<0.65)
plt.tight_layout()
In [ ]:
def segmentation_features(img, input_data, thresh):
  sq.im.process(img=img, layer="image", method="smooth")
  sq.im.segment(
      img=img,
      layer="image",
      method="watershed",
      channel=0,
      thresh=thresh, 
      geq=False
      )

  # define image layer to use for segmentation
  features_kwargs = {
      "segmentation": {
        "label_layer": "segmented_watershed",
        "props": ["label", "centroid"],
        "channels": [1, 2],
        }
      }

# calculate segmentation features
  sq.im.calculate_image_features(
      adata=input_data,
      img=img,
      layer="image",
      key_added="image_features",
      features_kwargs=features_kwargs,
      features="segmentation",
      mask_circle=True,
      n_jobs=2
      )
In [ ]:
segmentation_features(img_a, adata_tumor, 0.6)
segmentation_features(img_b, bdata_tumor, 0.58)
segmentation_features(img_c, cdata_tumor, 0.78)
segmentation_features(img_d, ddata_tumor, 0.65)

Processing image using `smooth` method
Adding image layer `image`
Finish (0:00:00)
Overwriting image layer `image_smooth`
Segmenting an image of shape `(1907, 2000, 1, 3)` using `SegmentationWatershed`
Adding image layer `image`
Finish (0:01:31)
Adding image layer `segmented_watershed`
Calculating features `['segmentation']` using `2` core(s)

Adding `adata.obsm['image_features']`
Finish (0:00:13)
Processing image using `smooth` method
Adding image layer `image`
Finish (0:00:00)
Overwriting image layer `image_smooth`
Segmenting an image of shape `(1789, 2000, 1, 3)` using `SegmentationWatershed`
Adding image layer `image`
Finish (0:01:01)
Adding image layer `segmented_watershed`
Calculating features `['segmentation']` using `2` core(s)

Adding `adata.obsm['image_features']`
Finish (0:00:04)
Processing image using `smooth` method
Adding image layer `image`
Finish (0:00:00)
Overwriting image layer `image_smooth`
Segmenting an image of shape `(1788, 2000, 1, 3)` using `SegmentationWatershed`
Adding image layer `image`
Finish (0:01:05)
Adding image layer `segmented_watershed`
Calculating features `['segmentation']` using `2` core(s)

Adding `adata.obsm['image_features']`
Finish (0:00:14)
Processing image using `smooth` method
Adding image layer `image`
Finish (0:00:00)
Overwriting image layer `image_smooth`
Segmenting an image of shape `(1978, 2000, 1, 3)` using `SegmentationWatershed`
Adding image layer `image`
Finish (0:01:29)
Adding image layer `segmented_watershed`
Calculating features `['segmentation']` using `2` core(s)

Adding `adata.obsm['image_features']`
Finish (0:00:07)
In [ ]:
adata_tumor.obs["cell_count"] = adata_tumor.obsm["image_features"]["segmentation_label"]
bdata_tumor.obs["cell_count"] = bdata_tumor.obsm["image_features"]["segmentation_label"]
cdata_tumor.obs["cell_count"] = cdata_tumor.obsm["image_features"]["segmentation_label"]
ddata_tumor.obs["cell_count"] = ddata_tumor.obsm["image_features"]["segmentation_label"]
In [ ]:
in_f = "oscc_sc_data.h5ad"
adata_sc = sc.read_h5ad(datadir + in_f)
sc.tl.rank_genes_groups(adata_sc, groupby="label", method="wilcoxon", use_raw=False)
markers_df = pd.DataFrame(adata_sc.uns["rank_genes_groups"]["names"]).iloc[0:100, :]
markers_df

ranking genes
    finished: added to `.uns['rank_genes_groups']`
    'names', sorted np.recarray to be indexed by group ids
    'scores', sorted np.recarray to be indexed by group ids
    'logfoldchanges', sorted np.recarray to be indexed by group ids
    'pvals', sorted np.recarray to be indexed by group ids
    'pvals_adj', sorted np.recarray to be indexed by group ids (0:00:10)
Out[ ]:
B cell CAF CRC DC EC MAF MSC Mac Mst Myo T cell
0 SSR4 SERPING1 KRT5 LAMP3 PECAM1 TAGLN NNMT HLA-DRA TPSB2 NEB PTPRC
1 MZB1 DCN KRT17 CCR7 S1PR1 TNS1 CHRNA1 HLA-DRB1 TPSAB1 ACTC1 TMSB4X
2 DERL3 C1R FXYD3 BIRC3 HYAL2 MYL9 RASD1 HLA-DQB1 CPA3 DES ARHGDIB
3 FKBP11 C1S PERP LY75 CDH5 ADAMTS1 CD63 CD74 TPSD1 FLNC STK17B
4 HERPUD1 COL1A2 DSP CD74 ENG CALD1 MEG3 HLA-DPA1 KIT H19 B2M
... ... ... ... ... ... ... ... ... ... ... ...
95 MALAT1 PRKCDBP TKT PAN3 COL4A1 MAP1B CLU CECR1 PAG1 APOBEC2 DAND5
96 ICAM2 TIMP3 ANXA8L1 TMSB4X PRSS23 EDNRA IP6K3 TLR2 CDK15 PADI2 L1TD1
97 RRBP1 DIO2 TNS4 HCK SPRY1 GSN MT1M SRGN MLPH MLIP CYTIP
98 RNF122 ARID5B SNRPE GLS DARC HSPB8 FOXO3 MSR1 LAX1 PGAM2 AKAP5
99 KLHL6 CD248 KRT6B STARD7 SCARF1 MOCS1 IFITM2 PPT1 RIN3 TBX15 AGMAT

100 rows × 11 columns

In [ ]:
genes_sc = np.unique(markers_df.melt().value.values)
genes_st = adata_tumor.var_names.values
markers = list(set(genes_sc).intersection(set(genes_st)))
len(markers)
Out[ ]:
338
In [ ]:
import torch
device = "cuda" if torch.cuda.is_available() else "cpu"
device
Out[ ]:
'cuda'
In [ ]:
data_ = None
data_ = bdata_tumor.copy()

ad_map = None
annotation_list = list()
df_all_genes = pd.DataFrame()
  
tg.pp_adatas(adata_sc, data_, genes=markers)

ad_map = tg.map_cells_to_space(adata_sc=adata_sc,     # scRNA-seq
                               adata_sp=data_,          # spatial data 
                               mode="constrained",
                               cluster_label='label',    # .obs field w cell types
                               target_count=data_.obs.cell_count.sum(),
                               density_prior=np.array(data_.obs.cell_count) / data_.obs.cell_count.sum(),
                               scale=True,
                               num_epochs=10000,
                               device=device,
                               learning_rate=0.1,
                               random_state=42
                               )

tg.project_cell_annotations(adata_map=ad_map, adata_sp=data_, annotation="label", threshold=0.5)
annotation_list = list(pd.unique(adata_sc.obs['label']))
data_.obs = pd.concat([data_.obs, data_.obsm["tangram_ct_pred"]], axis=1)

sc.pl.spatial(data_,
              color=annotation_list,
              vmax="p85",
              color_map="viridis",
              legend_fontsize=20,
              )

tg.plot_training_scores(ad_map, bins=20, alpha=.5)
ad_ge = tg.project_genes(adata_map=ad_map, adata_sc=adata_sc)
df_all_genes = tg.compare_spatial_geneexp(ad_ge, data_, adata_sc)
tg.plot_auc(df_all_genes)

tg.create_segment_cell_df(data_)
tg.count_cell_annotations(
    ad_map,
    adata_sc,
    data_,
    annotation="label",
    )

data_segment = tg.deconvolve_cell_annotations(data_)
sc.pl.spatial(
    data_segment[(data_segment.obs["cluster"] == "CRC" ) | (data_segment.obs["cluster"] == "CAF") | (data_segment.obs["cluster"] == "T cell")],
    color="cluster",
    size=0.05,
    show=True,
    frameon=True,
    alpha_img=0.05,
    legend_fontsize=20,
    )

plt.show()

Score: 0.692, KL reg: 0.038, Count reg: 1858.553, Lambda f reg: 1131.242
Score: 0.850, KL reg: 0.003, Count reg: 0.584, Lambda f reg: 202.166
Score: 0.852, KL reg: 0.003, Count reg: 0.060, Lambda f reg: 76.547
Score: 0.853, KL reg: 0.003, Count reg: 0.360, Lambda f reg: 57.193
Score: 0.853, KL reg: 0.003, Count reg: 0.210, Lambda f reg: 47.202
Score: 0.853, KL reg: 0.003, Count reg: 0.036, Lambda f reg: 39.972
Score: 0.853, KL reg: 0.003, Count reg: 0.180, Lambda f reg: 34.049
Score: 0.853, KL reg: 0.003, Count reg: 0.425, Lambda f reg: 30.705
Score: 0.853, KL reg: 0.003, Count reg: 0.052, Lambda f reg: 28.616
Score: 0.853, KL reg: 0.003, Count reg: 0.091, Lambda f reg: 26.902
Score: 0.853, KL reg: 0.003, Count reg: 0.663, Lambda f reg: 24.871
Score: 0.853, KL reg: 0.003, Count reg: 0.160, Lambda f reg: 22.332
Score: 0.853, KL reg: 0.003, Count reg: 0.803, Lambda f reg: 20.649
Score: 0.853, KL reg: 0.003, Count reg: 0.306, Lambda f reg: 18.362
Score: 0.853, KL reg: 0.003, Count reg: 0.467, Lambda f reg: 17.860
Score: 0.854, KL reg: 0.003, Count reg: 0.097, Lambda f reg: 18.176
Score: 0.854, KL reg: 0.003, Count reg: 0.453, Lambda f reg: 17.781
Score: 0.854, KL reg: 0.003, Count reg: 0.199, Lambda f reg: 17.061
Score: 0.854, KL reg: 0.003, Count reg: 0.689, Lambda f reg: 15.504
Score: 0.854, KL reg: 0.003, Count reg: 0.829, Lambda f reg: 14.351
Score: 0.854, KL reg: 0.003, Count reg: 0.044, Lambda f reg: 14.575
Score: 0.854, KL reg: 0.003, Count reg: 0.089, Lambda f reg: 13.723
Score: 0.854, KL reg: 0.003, Count reg: 0.622, Lambda f reg: 12.515
Score: 0.854, KL reg: 0.003, Count reg: 0.424, Lambda f reg: 12.695
Score: 0.854, KL reg: 0.003, Count reg: 0.191, Lambda f reg: 12.913
Score: 0.854, KL reg: 0.003, Count reg: 0.544, Lambda f reg: 12.545
Score: 0.854, KL reg: 0.003, Count reg: 0.668, Lambda f reg: 11.423
Score: 0.854, KL reg: 0.003, Count reg: 0.071, Lambda f reg: 10.677
Score: 0.854, KL reg: 0.003, Count reg: 0.522, Lambda f reg: 9.563
Score: 0.854, KL reg: 0.003, Count reg: 0.388, Lambda f reg: 9.688
Score: 0.854, KL reg: 0.003, Count reg: 0.224, Lambda f reg: 8.851
Score: 0.854, KL reg: 0.003, Count reg: 0.158, Lambda f reg: 8.867
Score: 0.854, KL reg: 0.003, Count reg: 0.298, Lambda f reg: 7.769
Score: 0.854, KL reg: 0.003, Count reg: 0.224, Lambda f reg: 7.837
Score: 0.854, KL reg: 0.003, Count reg: 0.144, Lambda f reg: 7.912
Score: 0.854, KL reg: 0.003, Count reg: 0.056, Lambda f reg: 7.994
Score: 0.854, KL reg: 0.003, Count reg: 0.187, Lambda f reg: 7.859
Score: 0.854, KL reg: 0.003, Count reg: 0.093, Lambda f reg: 7.949
Score: 0.854, KL reg: 0.003, Count reg: 0.183, Lambda f reg: 7.855
Score: 0.854, KL reg: 0.003, Count reg: 0.130, Lambda f reg: 7.905
Score: 0.854, KL reg: 0.003, Count reg: 0.197, Lambda f reg: 7.831
Score: 0.854, KL reg: 0.003, Count reg: 0.676, Lambda f reg: 6.359
Score: 0.854, KL reg: 0.003, Count reg: 0.611, Lambda f reg: 6.421
Score: 0.854, KL reg: 0.003, Count reg: 0.541, Lambda f reg: 6.489
Score: 0.854, KL reg: 0.003, Count reg: 0.461, Lambda f reg: 6.565
Score: 0.854, KL reg: 0.003, Count reg: 0.370, Lambda f reg: 6.654
Score: 0.854, KL reg: 0.003, Count reg: 0.043, Lambda f reg: 6.187
Score: 0.854, KL reg: 0.003, Count reg: 0.654, Lambda f reg: 5.371
Score: 0.854, KL reg: 0.003, Count reg: 0.601, Lambda f reg: 5.421
Score: 0.854, KL reg: 0.003, Count reg: 0.544, Lambda f reg: 5.477
Score: 0.854, KL reg: 0.003, Count reg: 0.480, Lambda f reg: 5.538
Score: 0.854, KL reg: 0.003, Count reg: 0.409, Lambda f reg: 5.608
Score: 0.854, KL reg: 0.003, Count reg: 0.313, Lambda f reg: 5.701
Score: 0.854, KL reg: 0.003, Count reg: 0.269, Lambda f reg: 4.749
Score: 0.854, KL reg: 0.003, Count reg: 0.220, Lambda f reg: 4.796
Score: 0.854, KL reg: 0.003, Count reg: 0.166, Lambda f reg: 4.849
Score: 0.854, KL reg: 0.003, Count reg: 0.105, Lambda f reg: 4.908
Score: 0.854, KL reg: 0.003, Count reg: 0.035, Lambda f reg: 4.976
Score: 0.854, KL reg: 0.003, Count reg: 0.214, Lambda f reg: 4.796
Score: 0.854, KL reg: 0.003, Count reg: 0.480, Lambda f reg: 3.533
Score: 0.854, KL reg: 0.003, Count reg: 0.454, Lambda f reg: 3.558
Score: 0.854, KL reg: 0.003, Count reg: 0.427, Lambda f reg: 3.585
Score: 0.854, KL reg: 0.003, Count reg: 0.396, Lambda f reg: 3.615
Score: 0.854, KL reg: 0.003, Count reg: 0.362, Lambda f reg: 3.647
Score: 0.854, KL reg: 0.003, Count reg: 0.326, Lambda f reg: 3.683
Score: 0.854, KL reg: 0.003, Count reg: 0.287, Lambda f reg: 3.722
Score: 0.854, KL reg: 0.003, Count reg: 0.243, Lambda f reg: 3.764
Score: 0.854, KL reg: 0.003, Count reg: 0.195, Lambda f reg: 3.812
Score: 0.854, KL reg: 0.003, Count reg: 0.139, Lambda f reg: 3.867
Score: 0.854, KL reg: 0.003, Count reg: 0.020, Lambda f reg: 4.014
Score: 0.854, KL reg: 0.003, Count reg: 0.705, Lambda f reg: 2.303
Score: 0.854, KL reg: 0.003, Count reg: 0.695, Lambda f reg: 2.313
Score: 0.854, KL reg: 0.003, Count reg: 0.682, Lambda f reg: 2.325
Score: 0.854, KL reg: 0.003, Count reg: 0.667, Lambda f reg: 2.339
Score: 0.854, KL reg: 0.003, Count reg: 0.653, Lambda f reg: 2.353
Score: 0.854, KL reg: 0.003, Count reg: 0.637, Lambda f reg: 2.369
Score: 0.854, KL reg: 0.003, Count reg: 0.620, Lambda f reg: 2.386
Score: 0.854, KL reg: 0.003, Count reg: 0.601, Lambda f reg: 2.404
Score: 0.854, KL reg: 0.003, Count reg: 0.581, Lambda f reg: 2.424
Score: 0.854, KL reg: 0.003, Count reg: 0.559, Lambda f reg: 2.445
Score: 0.854, KL reg: 0.003, Count reg: 0.536, Lambda f reg: 2.468
Score: 0.854, KL reg: 0.003, Count reg: 0.510, Lambda f reg: 2.494
Score: 0.854, KL reg: 0.003, Count reg: 0.482, Lambda f reg: 2.521
Score: 0.854, KL reg: 0.003, Count reg: 0.451, Lambda f reg: 2.552
Score: 0.854, KL reg: 0.003, Count reg: 0.418, Lambda f reg: 2.585
Score: 0.854, KL reg: 0.003, Count reg: 0.380, Lambda f reg: 2.622
Score: 0.854, KL reg: 0.003, Count reg: 0.339, Lambda f reg: 2.664
Score: 0.854, KL reg: 0.003, Count reg: 0.291, Lambda f reg: 2.711
Score: 0.854, KL reg: 0.003, Count reg: 0.235, Lambda f reg: 2.767
Score: 0.854, KL reg: 0.003, Count reg: 0.078, Lambda f reg: 2.014
Score: 0.854, KL reg: 0.003, Count reg: 0.514, Lambda f reg: 1.490
Score: 0.854, KL reg: 0.003, Count reg: 0.504, Lambda f reg: 1.499
Score: 0.854, KL reg: 0.003, Count reg: 0.495, Lambda f reg: 1.508
Score: 0.854, KL reg: 0.003, Count reg: 0.484, Lambda f reg: 1.518
Score: 0.854, KL reg: 0.003, Count reg: 0.474, Lambda f reg: 1.529
Score: 0.854, KL reg: 0.003, Count reg: 0.462, Lambda f reg: 1.540
Score: 0.854, KL reg: 0.003, Count reg: 0.449, Lambda f reg: 1.553
Score: 0.854, KL reg: 0.003, Count reg: 0.436, Lambda f reg: 1.566
Score: 0.854, KL reg: 0.003, Count reg: 0.421, Lambda f reg: 1.581
Score: 0.854, KL reg: 0.003, Count reg: 0.404, Lambda f reg: 1.597

<Figure size 800x640 with 0 Axes>