-
Notifications
You must be signed in to change notification settings - Fork 1
/
Snakefile.py
204 lines (179 loc) · 7.44 KB
/
Snakefile.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
import os
import sys
import pandas as pd
import numpy as np
from itertools import product
OUTDIR = os.path.abspath(config['outdir'])
SRCDIR = os.path.join(os.path.dirname(os.path.abspath(workflow.snakefile)), 'src')
LOGDIR = os.path.join(OUTDIR, 'logs')
FLEXYNESIS = config['flexynesis']
def get_data_url(task):
url = config['tasks'][task]['url']
base = os.path.basename(url)
name = os.path.splitext(base)[0]
return url, base, name
def get_data_path(task_df, prefix):
task = task_df[task_df['prefix'] == prefix]['task'].item()
return os.path.join(OUTDIR, "data", task, ".dummy")
def parse_vars(s):
# Improved with handling for spaces within the values
return dict(pair.split(':') for pair in s.split())
def parse_tool(tool_string):
"""
Parses a tool string to identify the tool and its optional convolution type.
Returns the tool and conv_type (None if not specified).
"""
parts = tool_string.split(':', 1)
return (parts[0], parts[1]) if len(parts) > 1 else (parts[0], None)
def get_combinations(task_settings):
# Unpack settings
variables = [parse_vars(s) for s in task_settings['vars']]
tools = task_settings['tools'].strip().split(',')
data_types = task_settings['data_types']
fusions = task_settings['fusions']
loss_weighting = ['True'] #, 'False'] # try both setting
# single switches
min_features = task_settings['min_features']
hpo_iterations = task_settings['hpo_iterations']
early_stop_patience = task_settings['early_stop_patience']
log_transform = task_settings['log_transform']
features_top_percentile = task_settings['features_top_percentile']
finetuning = [int(x.strip()) for x in task_settings['finetuningSampleN'].split(',')]
# Generate all combinations
combinations = product(variables, tools, fusions, data_types, loss_weighting, finetuning)
combs = []
for variables, tool, fusion, data_type, loss_weighting, finetuning in combinations:
tool, gnn_conv = parse_tool(tool)
# handle some exceptions
if tool in ['GNN', 'RandomForest', 'SVM', 'RandomSurvivalForest']:
fusion = 'early' # these tools supports early fusion only
# if single data modality, set fusion to early
if len(data_type.strip().split(',')) == 1:
# multimodal
fusion = 'early'
if tool in ['RandomForest', 'SVM', 'RandomSurvivalForest']:
finetuning = 0 #these tools don't support finetuning
arguments = {
'task': task,
'tool': tool,
'gnn_conv': gnn_conv, #optional
'target': variables.get('target', None),
'batch': variables.get('batch', None),
'event': variables.get('event', None),
'time': variables.get('time', None),
'data_types': data_type,
'fusion': fusion,
'hpo_iter': hpo_iterations,
'early_stop_patience': early_stop_patience,
'features_min': min_features,
'feature_perc': features_top_percentile,
'log_transform': log_transform,
'use_loss_weighting': loss_weighting,
'finetuning_samples': finetuning
}
combs.append(arguments)
return combs
def get_model_args(task_df, prefix):
# Retrieve the row for the given prefix to minimize repetitive indexing
task_row = task_df[task_df['prefix'] == prefix].iloc[0]
datapath = os.path.dirname(get_data_path(task_df, prefix))
args = [
"--data_path", datapath,
"--model_class", task_row['tool'],
"--fusion_type", task_row['fusion'],
"--hpo_iter", str(task_row['hpo_iter']),
"--early_stop_patience", str(task_row['early_stop_patience']),
"--features_min", str(task_row['features_min']),
"--features_top_percentile", str(task_row['feature_perc']),
"--use_loss_weighting", task_row['use_loss_weighting'],
"--data_types", task_row['data_types'],
"--log_transform", str(task_row['log_transform']),
"--finetuning_samples", str(task_row['finetuning_samples'])
]
if task_row['target']:
args.extend(["--target_variables", task_row['target']])
if task_row['batch']:
args.extend(["--batch_variables", task_row['batch']])
if task_row['event']:
args.extend(["--surv_event_var", task_row['event']])
if task_row['time']:
args.extend(["--surv_time_var", task_row['time']])
if task_row['gnn_conv']:
args.extend(["--gnn_conv_type", task_row['gnn_conv']])
# Join the arguments into a command-line friendly string
command_line_args = " ".join(args)
return command_line_args
targets = []
for task, settings in config['tasks'].items():
targets.extend(get_combinations(settings))
task_df = pd.DataFrame(targets).drop_duplicates().reset_index(drop=True)
task_df['prefix'] = ['analysis' + str(x) for x in task_df.index]
print(task_df)
TASKS=list(np.unique(task_df['task']))
ANALYSES=list(task_df['prefix'])
rule all:
input:
# print analysis table
os.path.join(OUTDIR, "analysis_table.csv"),
# input data download
expand(os.path.join(OUTDIR, "data", "{task}", ".dummy"), task = TASKS),
# modeling results
expand(os.path.join(OUTDIR, "results", "{analysis}.{output_type}.csv"),
analysis = ANALYSES,
output_type = ['stats']),
# dashboard
os.path.join(OUTDIR, "dashboard.html")
rule print_analysis_table:
output:
os.path.join(OUTDIR, "analysis_table.csv")
run:
task_df.to_csv(output[0])
rule download_data:
input:
os.path.join(OUTDIR, "analysis_table.csv") # make sure analysis table is printed first.
output:
os.path.join(OUTDIR, "data", "{task}", ".dummy")
log:
os.path.join(LOGDIR, "download.{task}.log")
params:
url = lambda wildcards: get_data_url(wildcards.task)[0],
# download to task folder
downloaded_tgz = lambda wildcards: os.path.join(OUTDIR, "data", get_data_url(wildcards.task)[1]),
# extract the folder
extracted = lambda wildcards: get_data_url(wildcards.task)[2],
# name folder to "task" name
new_name = lambda wildcards: os.path.join(OUTDIR, "data", wildcards.task),
shell:
"""
curl -L -o {params.downloaded_tgz} {params.url}
tar -xzvf {params.downloaded_tgz}
mv {params.extracted}/* {params.new_name}; rm -rf {params.extracted}
touch {output}
"""
rule model:
input:
lambda wildcards: get_data_path(task_df, wildcards.analysis)
output:
os.path.join(OUTDIR, "results", "{analysis}.stats.csv")
log:
os.path.join(LOGDIR, "{analysis}.log")
params:
args = lambda wildcards: get_model_args(task_df, wildcards.analysis),
outdir = os.path.join(OUTDIR, "results")
shell:
"""
{FLEXYNESIS} {params.args} --outdir {params.outdir} --prefix {wildcards.analysis} > {log} 2>&1
"""
rule dashboard:
input:
expand(os.path.join(OUTDIR, "results", "{analysis}.{output_type}.csv"),
analysis = ANALYSES,
output_type = ['stats'])
output:
os.path.join(OUTDIR, "dashboard.html")
log:
os.path.join(LOGDIR, "dashboard.log")
params:
dashboard_script = os.path.join(SRCDIR, "main.py"),
shell:
"python {params.dashboard_script} {OUTDIR} {output} > {log} 2>&1"