Skip to content

Commit

Permalink
[USGS-R#184] plots that show the 'holdout' performance
Browse files Browse the repository at this point in the history
  • Loading branch information
@jsadler2
jsadler2 committed Sep 1, 2023
1 parent b5bea1b commit 5f305d6
Showing 1 changed file with 195 additions and 39 deletions.
234 changes: 195 additions & 39 deletions 3_visualize/src/python_scripts/plot_pred_performance.py
View file
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
# ---
# jupyter:
# jupytext:
# formats: ipynb,py:light
# text_representation:
# extension: .py
# format_name: light
Expand All @@ -13,65 +14,220 @@
# ---

import pandas as pd
import os
import seaborn as sns
import matplotlib.pyplot as plt
import plot_utils
from plot_utils import read_and_filter_df
import numpy as np
import seaborn.objects as so


# +
run_id = 215

def read_and_filter_df(metric_type, partition):
f_name = f"../../../2a_model/out/models/combined_{metric_type}_metrics.csv"
df_comb = pd.read_csv(f_name, dtype={"site_id": str})
df_comb = df_comb[df_comb['partition'] == partition]
df_comb = df_comb[df_comb['variable'].str.startswith('do')]
df_comb = df_comb[df_comb['rmse'].notna()]
return df_comb
outdir = f"../../out/pred_perf/{run_id}"

if not os.path.exists(outdir):
os.makedirs(outdir)
# -


models = [
"0_baseline_LSTM",
"1_metab_multitask",
"1a_multitask_gpp_er",
"2_multitask_dense",
]

df_comb_reach = read_and_filter_df("reach", "val")
df_comb_reach = df_comb_reach.replace(
{
"0_baseline_LSTM": "Baseline",
"1a_multitask_do_gpp_er": "Metab Multitask - GPP, ER",
"1_metab_multitask": "Metab Multitask",
"2_multitask_dense": "Metab Dependent",
}
)

######## Barplot by site ######################################################
g = sns.catplot(x='site_id', y='rmse', col='variable', data=df_comb_reach,
hue='model_id', kind='bar', legend=False, ci='sd',
hue_order=plot_utils.model_order)
g.set_xticklabels(rotation=90)
for ax in g.axes.flatten():
ax.grid()
ax.set_axisbelow(True)
plot_utils.mark_val_sites(ax)
df_comb_reach.holdout.unique()


test_sites_urban = ["01475530", "01475548"]


def define_group(row):
if row["holdout"] != "temporal" and row["holdout"] != "1_urban":
return "spatial non-urban"
elif row["holdout"] == "temporal":
if row['site_id'] in test_sites_urban:
return "temporal urban"
else:
return "temporal non-urban"
elif row["holdout"] == '1_urban':
return "spatial one-urban"


df_comb_reach["holdout_id"] = df_comb_reach.apply(define_group, axis=1)

df_comb_reach.holdout_id.unique()


def plot_by_site_or_holdout(data, x, kind, outfile):
g = sns.catplot(
x=x,
y="rmse",
col="variable",
data=data,
hue="model_id",
kind=kind,
legend=False,
errorbar="sd",
col_order=["do_min", "do_max"],
dodge=True
)
g.set_xticklabels(rotation=90)

for i, ax in enumerate(g.axes.flatten()):
ax.grid()
ax.set_axisbelow(True)

g.axes.flatten()[0].set_ylabel("RMSE (mg O2/l)")
ax.legend(bbox_to_anchor=(1.05, 0.55))
plt.tight_layout()
plt.show()
plt.savefig(os.path.join(outdir, outfile), dpi=300)
plt.clf()



######## Barplot by site (temporal)############################################
df_comb_reach_temporal = df_comb_reach[df_comb_reach['holdout_id'] == 'temporal non-urban']
plot_by_site_or_holdout(df_comb_reach_temporal, "site_id", "bar", "val_results_by_site.jpg")



######## stripplot by site (temporal)############################################
df_comb_reach_temporal = df_comb_reach[df_comb_reach['holdout_id'] == 'temporal non-urban']
plot_by_site_or_holdout(df_comb_reach_temporal, "site_id", "strip", "val_results_by_site_strip.jpg")


######## stripplot by site (temporal)############################################
df_comb_reach_spatial = df_comb_reach[df_comb_reach['holdout_id'] == 'spatial non-urban']
plot_by_site_or_holdout(df_comb_reach_spatial, "site_id", "strip", "val_results_by_site_strip.jpg")


######## Barplot by holdout ######################################################
plot_by_site_or_holdout(df_comb_reach, "holdout_id", "bar", "val_results_by_holdout.jpg")

plt.legend(bbox_to_anchor=(1.05, .55))

# +
######## Stripplot by holdout ####################################################
plot_by_site_or_holdout(df_comb_reach, "holdout_id", "strip", "val_results_by_holdout_strip.jpg")


######## Barplot overall ######################################################
df_comb = read_and_filter_df("overall", "val")

# +
fig, ax = plt.subplots(figsize=(17, 6))
ax = sns.barplot(
x="variable", y="rmse", data=df_comb, hue="model_id", ax=ax
) # , hue_order=['0_baseline_LSTM', '1a_multitask_do_gpp_er', '1_metab_multitask', '2_multitask_dense'])
for c in ax.containers:
ax.bar_label(c, label_type="center", fmt="%.2f")

ax.set_xlabel("")
plt.legend(loc="lower left", bbox_to_anchor=(1.05, 0), title="Model")
plt.tight_layout()
plt.savefig("../../out/pred_perf/val_results_by_site.png")
plt.clf()
plt.savefig(os.path.join(outdir, "val_results_overall.png"), dpi=300)

######## Stripplot by site ####################################################
g=sns.catplot(x='site_id', y='rmse', hue='model_id', col='variable',
col_wrap=3, data=df_comb_reach, dodge=True, legend=False,
hue_order=plot_utils.model_order)
g.set_xticklabels(rotation=90)
g.set_titles('{col_name}')
for site_id, ax in g.axes_dict.items():
ax.grid()
plot_utils.mark_val_sites(ax)

plt.legend(bbox_to_anchor=(1.05, .55))
######## Barplot calculating site metrics then averaging #########################################
fig, ax = plt.subplots(figsize=(15, 4))
ax = sns.barplot(
x="variable",
y="rmse",
hue="model_id",
data=df_comb_reach,
order=["do_min", "do_max"],
)

for c in ax.containers:
ax.bar_label(c, label_type="center", fmt="%.2f")

ax.set_ylabel("RMSE (mg O2/l)")

plt.legend(loc="lower left", bbox_to_anchor=(1.05, 0), title="Model")
plt.tight_layout()
plt.savefig('../../out/pred_perf/val_results_by_site_strip.png')
plt.clf()
plt.savefig(os.path.join(outdir, "val_results_overall_avg_across_sites.jpg"), dpi=300)


######## Barplot overall ######################################################
df_comb = read_and_filter_df('overall', 'val')
# +
######## Barplot calculating site metrics then median-ing #########################################
fig, ax = plt.subplots(figsize=(17, 3))
ax = sns.barplot(
x="variable",
y="rmse",
hue="model_id",
data=df_comb_reach,
hue_order=None,
estimator=np.median,
)

fig, ax = plt.subplots(figsize=(6,4))
ax = sns.barplot(x='variable', y='rmse', data=df_comb, hue='model_id', ax=ax,
hue_order=plot_utils.model_order)
for c in ax.containers:
ax.bar_label(c, label_type="center", fmt='%.2f')
ax.bar_label(c, label_type="center", fmt="%.2f")

plt.legend(loc="lower left", bbox_to_anchor=(1.05, 0), title="Model")
plt.savefig(os.path.join(outdir, "val_results_overall_median_across_sites.jpg"), dpi=300)
# -

df_comb_month = read_and_filter_df("month", "val")

month_order = [9, 10, 11, 12, 1, 2, 3, 4, 5, 6, 7, 8]

# +
######## Barplot by month #########################################
g = sns.catplot(
x="date",
y="rmse",
col="variable",
data=df_comb_month,
hue="model_id",
kind="strip",
legend=False,
ci="sd",
dodge=True,
# hue_order=models,
)

ax.set_xlabel('')
for ax in g.axes.flatten():
ax.grid()
ax.set_axisbelow(True)
g.set_xlabels("month")
plt.legend(bbox_to_anchor=(1.05, 0.55))
plt.tight_layout()
plt.savefig(os.path.join(outdir, "val_results_by_month_strip.jpg"), dpi=300)
plt.clf()
# -

plt.savefig("../../out/pred_perf/val_results_overall.png")
df_comb_month.sort_values("rmse").iloc[-1]

df_comb_month[df_comb_month["date"] == 12].sort_values("rmse").iloc[-1]

df_2 = df_comb_month[
(df_comb_month["rep_id"] == 2) & (df_comb_month["model_id"].str.startswith("2"))
]
df_2_pg = df_comb_month[df_comb_month["model_id"].str.startswith("2")]

sns.barplot(x="date", y="rmse", hue="variable", data=df_2)

g = sns.catplot(
x="date", y="rmse", hue="rep_id", col="variable", data=df_2_pg, col_wrap=1, aspect=2
)
for ax in g.axes.flatten():
ax.grid()
ax.set_axisbelow(True)
g.set_xlabels("month")
plt.tight_layout()
plt.savefig(os.path.join(outdir, "2_monthly_performance.jpg"), dpi=300)

0 comments on commit 5f305d6

Please sign in to comment.